Allocation of human capital and innovation at the …...Allocation of human capital and innovation at the frontier: Firm-level evidence on Germany and the Netherlands Eric BARTELSMANy

Allocation of human capital and innovation at the frontier:Firm-level evidence on Germany and the Netherlands∗

Eric BARTELSMAN† Sabien DOBBELAERE‡ Bettina PETERS§

July, 2013

Abstract

This paper examines how productivity effects of human capital and innovation vary at different pointsof the conditional productivity distribution. Our analysis draws upon two large unbalanced panels of 6,634enterprises in Germany and 14,586 enterprises in the Netherlands over the period 2000-2008, considering5 manufacturing and services industries that differ in the level of technological intensity. Industries inthe Netherlands are characterized by a larger average proportion of high-skilled employees and indus-tries in Germany by a more unequal distribution of human capital intensity. Except for low-technologymanufacturing, average innovation performance is higher in all industries in Germany and the innova-tion performance distributions are more dispersed in the Netherlands. In both countries, we observenon-linearities in the productivity effects of investing in product innovation in the majority of indus-tries. Frontier firms enjoy the highest returns to product innovation whereas the most negative returns toprocess innovation are observed in the best-performing enterprises of most industries. In both countries,we find that the returns to human capital increase with proximity to the technological frontier in industrieswith a low level of technological intensity. Strikingly, a negative complementarity effect between humancapital and proximity to the technological frontier is observed in knowledge-intensive services, which ismost pronounced for the Netherlands. Suggestive evidence for the latter points to a winner-takes-allinterpretation of this finding.

JEL classification : C10, I20, O14, O30.Keywords : Human capital, innovation, productivity, quantile regression.

1 Introduction

Over the past decade, concerns about the observed poor productivity performance in European countries,in particular compared to the US, have been expressed not only in academic circles but also among policymakers and politicians. Since the mid 1990s, the productivity gap between Europe and the US has risendramatically: GDP per hour worked in the EU has decreased from 98.3 percent of the US level in 1995 to

∗Financial support from the SEEK (Strenghtening Effi ciency and Competitiveness in the European Knowledge Economies)research program is gratefully acknowledged. The authors would like to thank Statistics Netherlands for providing the Dutchdata and are grateful to Alex Coad and participants at conferences and seminars for helpful comments and suggestions.†VU University Amsterdam, Tinbergen Institute and IZA.‡VU University Amsterdam, Tinbergen Institute and IZA.§Centre for European Economic Research (ZEW), MaCCI Mannheimer Centre for Competition and Innovation and University

of Zurich.

1

90.0 percent in 2006. Numerous reports view Europe’s unsatisfactory growth performance as a signal of itsfailure to transform into a knowledge-based economy (Kok, 2004; Sapir et al., 2004; European Commission,2008). Research into the policy drivers of a knowledge-based economy has taken many disparate routes,from theoretical modeling using an aggregate (macro) perspective to empirical explorations using firm-level(micro) data. Starting from the observation that neither micro evidence, nor meso evidence per se conclusivelyidentifies the drivers that boost productivity, this paper takes an integrated micro-meso approach to examinethe role of innovation and human capital in shaping the productivity distribution.

We first investigate firm-level heterogeneity in the productivity effects of the strategies to invest in innovationand human capital. Motivated by the increasing prominence of services in European countries and the centralrole played by knowledge-intensive services in the knowledge-based economies, we focus on high- versus low-technology manufacturing and services industries in Germany and the Netherlands. Given that (i) evenwithin these industries, there is significant heterogeneity between firms and (ii) the returns to innovation andhuman capital are highly skewed, we use quantile regression techniques to study the relationship betweeninnovative activity and human capital on the one hand and productivity on the other hand at different pointsof the conditional productivity distribution. In a subsequent, more descriptive step, we exploit the degree ofheterogeneity in the returns to innovation and human capital to re-examine differences in the productivitydistribution between industries.

There is a vast empirical literature on the effect of investments in R&D and innovation on productivity. Onaverage, the private returns to R&D are strongly positive and somewhat higher than for ordinary capital (seeHall et al., 2010 for a survey). On average, there are substantial positive impacts of product innovation onrevenue productivity whilst the impact of process innovation is more ambiguous (see Hall, 2011 for a survey).Existing quantile regression analyses provide mixed evidence on the relationship between different aspects ofR&D and innovation on the one hand and performance on the other hand (see e.g. include Coad and Rao,2006; Coad and Rao, 2008; Ebersberger et al., 2010; Falk, 2012; Goedhuys and Sleuwaegen, 2009; Hölzl,2009; Kaiser, 2009; Love et al., 2009; Mata and Wörter, 2013; Nahm, 2001; Segarra and Teruel, 2011 andStam and Wennberg, 2009).

Likewise, there is a vast empirical literature on the effects of human capital on productivity. On average,human capital returns are found to be significantly positive, particularly in recent studies on the impact ofhuman capital on productivity (see e.g. Fox and Smeets, 2011 and Mason et al., 2012).1 Using a large panelof all Danish citizens over the period 1980-2001, Fox and Smeets (2011) construct detailed firm-level humancapital measures and find that human capital inputs raise firm output considerably. Using cross-countryindustry-level data for 26 industries in 5 countries (France, Germany, the Netherlands UK and US) over theperiod 1979-2000, Mason et al. (2012) provide evidence of positive human capital returns, particularly whenusing a composite human capital variable accounting for both certified skills (educational attainment) and

1Up to the first half of the nineties, macro evidence pointed to a positive relationship between human capital and outputgrowth. However, this evidence was refuted by subsequent studies during the second half of the nineties (see Sianesi and vanReenen, 2003 and de la Fuente, 2011 for a survey). The latter finding is largely explained by methodological diffi culties relatedto measuring skills and modeling the channels through which skills impact on economic performance. Starting with Kruegerand Lindhal (2001), considerable progress has been made to tackle these methodological problems (see e.g. Cohen and Soto,2007; de la Fuente and Doménech, 2001, 2006; Barro and Lee, 2010). As a results, the latter studies find again positive impactsof education on economic growth. Another set of recent studies, focusing on the quality of education rather than its quantity,show even larger productivity effects (e.g. Coulombe, 2004; Hanushek and Kimko, 2000; Hanushek and Wößmann, 2008, 2012).

2

uncertified skills acquired through on-the-job training and experience. To our knowledge, our study is thefirst to examine human capital returns at different points of the conditional productivity distribution.2

Our analysis draws upon specific elements of recent endogenous growth models confirming that economy-wide technological improvements occur through the channel of innovation in advanced economies. Benhabiband Spiegel (1994), Acemoglu et al. (2003, 2006) and Vandenbussche et al. (2006) share the underlying ideathat technological improvements are the result of a combination of innovation and imitation. In particular,Acemoglu et al. (2003, 2006) show that innovation becomes more important than imitation as an economicentity approaches the technological frontier. Inspired by the argument of Nelson and Phelps (1966) thateducation facilitates the implementation of new technologies and adapting their framework to allow for thecatch-up of technology to the technology of the leading country, Benhabib and Spiegel (1994) provide cross-country evidence that countries with higher education tend to close the technological gap faster than othersand experience higher economic growth. Vandenbussche et al. (2006) go one step further and show that thecontribution of human capital to productivity growth can be decomposed into a level effect and a compositioneffect. In line with Acemoglu (2006), they assume that unskilled labor is better suited to imitation whereasmore intensive use of skilled labor is required for innovation. Taking into account endogenous labor allocationacross these imitation and innovation activities, they argue that one needs to account for both an economy’sdistance to the technological frontier and the composition of its human capital, which they empirically confirmat the macro level.

In a broad sense, our study fits into the empirical literature advocating that growth-maximizing policiesshould depend on the distance to the technological frontier. So far, the role of distance to the technologicalfrontier has predominantly been examined using industry-level data. Existing articles include Griffi th etal. (2003, 2004) on R&D intensity in a panel of industries across 12 OECD countries, Aghion et al. (2004)on threat of entry in UK industries, Aghion et al. (2005) on product market competition in UK industries,Kneller and Stevens (2006) on human capital and R&D in a panel of industries across 12 OECD countries,Aghion et al. (2008) on the liberalization of product entry in India and Acemoglu et al. (2006) on opennessto trade, entry costs and schooling level in a cross-country panel of about 100 non-OECD countries. In amore narrow sense, our study is most closely related to Vandenbussche et al. (2006), Inklaar et al. (2008)and Madsen et al. (2010). Using a panel dataset covering 19 OECD countries between 1960 and 2000,Vandenbussche et al. (2006) provide evidence of skilled labor having a higher growth-enhancing effect closerto the technological frontier. Using EUKLEMS industry data on multifactor productivity covering the period1995-2004, Inklaar et al. (2008), however, do not find support for the argument that there are productivityexternalities from employing university-educated workers for leaders in market services industries. Using apanel of 23 OECD countries and 32 developing countries covering the period 1970-2004, Madsen et al. (2010)show that R&D intensity, its interaction with distance to the frontier, educational attainment interacted withdistance to the frontier and technological gap influence total factor growth positively and point to differenteffects for developed versus developing countries.

A detailed look at our data reveals three stylized facts about human capital, innovation and productivity inGermany and the Netherlands. First, irrespective of their level of technological intensity, industries in theNetherlands are characterized by a higher average share of employees possessing a college or university degree

2Existing quantile regression studies have focused on changes in the returns of skills at different points of the wages/earningsdistribution (see e.g. Arias et al., 2001; Buchinsky, 1994; Buchinsky, 2001; Chevalier et al., 2004; Choi and Jeong, 2007; Dennyand O’Sullivan, 2007; Flabbi et al., 2008; Harmon et al., 2003; Hartog et al., 2001; Machado and Mata, 2001; Machado andMata, 2005; Martins and Pereira, 2004; Mwabu-Schultz, 1996; Pereira and Martins, 2002 and Tobias, 2002).

3

and industries in Germany by a more unequal distribution of human capital intensity. Second, average inno-vation performance —measured by the logarithm of real innovative sales per employee for product innovators—is higher in all industries, except for Low-technology manufacturing in Germany and the innovation perfor-mance distributions are more dispersed in all Dutch industries, except for Low-technology manufacturing.Third, average productivity is higher in all manufacturing industries in the Netherlands and productivity ismore unequally distributed in all industries, except for High-technology manufacturing in Germany.

Allowing the productivity effects of human capital and innovation to vary at different points of the conditionalproductivity distribution, our two main findings are summarized as follows. First, we find increasing marginalreturns to product innovation as we move up through the productivity distribution but negative marginalreturns to process innovation for the best-performing enterprises in the majority of industries in both coun-tries. Apparently, the best strategy for frontier firms is to focus on product rather than on process innovation.Second, the returns to human capital increase with proximity to the technological frontier in industries witha low level of technological intensity in both countries, thereby providing micro-evidence on the positivecomplementarity effect put forward by Vandenbussche et al. (2006). Strikingly, we find a negative comple-mentarity effect between human capital and proximity to the technological frontier in knowledge-intensiveservices. The latter finding is most pronounced for the Netherlands.3 We provide suggestive evidence insupport of a winner-takes-all interpretation for the Netherlands. Investment in intangibles in knowledge-intensive services, making use of human capital intensely, might lead to a profitable breakthrough for onefirm which could compensate the losses of many competitors.

We proceed as follows. Section 2 elucidates our empirical strategy. Section 3 discusses the data for Germanyand the Netherlands. Section 4 presents some stylized facts. Section 5 reports the results. Section 6 concludes.

2 Econometric framework

Our empirical analysis consists of two parts. In the first part, we estimate the returns to human capital andinnovation at the firm level. Our econometric framework is based on an augmented standard Cobb-Douglasproduction function approach. The logarithmic specification of the production function in intensity form isgiven by:

ln

(Q

L

)it

= β0 + βK ln

(K

L

)it

+ βM ln

(M

L

)it

+ βL lnLit (1)

+βHCHCit + βCTFCTFit−1 + βPDPDit + βPCPCit + αControls+ uit

where Qit is output of firm i in year t, and L, K andM denote the number of employees, physical capital andmaterial, respectively. Although productivity is measured in intensity form, firm size (lnL) is additionallyincluded. It allows to test for the hypothesis of constant returns to scale which corresponds to βL = 0. Theproduction function is extended by including human capital (HC), the technological position of the firm(CTF ), product innovation (PD) and process innovation (PC). We further account for the productivityimpact of some additional control variables (Controls) that will be explained in more detail in Section 3.3.βK and βM measure the output elasticity of capital and material whilst βHC , βPD and βPC capture thereturns to human capital, product and process innovation respectively.

3The latter result no longer holds for Germany in regressions that also estimate individual fixed effects.

4

We estimate this production function at the country-industry level using four different estimation methodsthat differ in their degree of firm-level heterogeneity they account for. Standard least squares regressiontechniques (OLS) provide point estimates for the average productivity effect of the independent variablesin a ‘representative enterprise’. Unobserved heterogeneity among firms, however, may make it diffi cult toisolate the productivity effects of human capital and innovation as both variables are likely to correlate withunobserved firm characteristics such as managerial ability. As an additional source of heterogeneity, wetherefore account for firm-specific effects in estimating the average returns to human capital and innovationby using the fixed effects (FE) estimator.

The exclusive focus on mean effects of OLS and FE may be misleading in our study since it seems unlikelythat most firms obtain the ‘average’return to human capital and innovation or even close to it. In order toobtain a more detailed picture of heterogenous returns, we therefore use quantile regression (QR) techniquesto model the conditional productivity distribution at various quantiles θ, conditional on the explanatoryvariables. The use of quantile regression techniques provides two other major advantages. First, whilst theoptimal properties of standard regression estimators are not robust to modest departures from normality,quantile regression results are characteristically robust to outliers and heavy-tailed distributions. In fact, thequantile regression solution is invariant to outliers of the dependent variable that tend to ±∞ (Buchinsky,1994). Second, a quantile regression approach avoids the restrictive assumption that the error terms areidentically distributed at all points of the conditional distribution.

The quantile regression model, first introduced in Koenker and Bassett’s (1978) seminal contribution, can bewritten as:

yit = x′

itβθ + uθit with Qθ (yit|xit) = x′

itβθ (2)

where yit is the dependent variable, xit a (K × 1)-vector of regressors, βθ the (K × 1)-vector of parametersto be estimated and uθit the error term. Qθ (yit|xit) denotes the θth conditional quantile of yit given xit, with0 < θ < 1. The θth conditional quantile function can be estimated by solving the following minimizationproblem:

minβ

1

n

∑i,t:yit≥x′itβ

θ|yit − x′

itβ|+∑

i,t:yit<x′itβ

(1− θ) |yit − x′

itβ|

= minβ

1

n

∑ρθuθit (3)

where ρθuθit, known as the ‘check function’, is defined as

ρθuθit =

{θuθit if uθit ≥ 0(θ − 1)uθit if uθit < 0

(4)

Eq. (3) is solved by linear programming methods. As one increases θ continuously from 0 to 1, one traces theentire conditional distribution of y, conditional on x (Buchinsky, 1994). In our study, the parameter estimatefor the kth exogenous variable, let’s say human capital, is interpreted as the marginal change in productivitydue to a marginal change in human capital conditional on being on the θth quantile of the distribution. Thisis also called the θth quantile return to human capital. We are particularly interested in how these returnschange along the distribution.

The standard quantile regression method allows the impact of all explanatory variables to vary along the con-ditional productivity distribution. However, it does not account for other unobserved firm-specific variables

5

αi that might affect productivity. In order to take this type of heterogeneity into account when estimatingthe quantile returns to human capital and innovation, we use a fixed effects quantile regression (FEQR)estimator for panel data (Koenker, 2005):

yit = x′

itβθ + αi + εθit (5)

We employ a two-step estimation approach. In the first step, we estimate a standard fixed effects model andpredict the firm-specific effects α̂i = yi − xi′β̂FE . In the second step, we run a pooled quantile regression ofyit − α̂i on all explanatory variables xit in order to obtain quantile regression estimates for βθ.

Based on our firm-level results, we examine in the second part of our empirical analysis whether heterogenousproductivity effects of human capital and innovation significantly change productivity distribution charac-teristics at the industry level. We follow an approach proposed by Machado and Mata (2000) and recentlyused by Mata and Wörter (2013) to investigate the effect of internal and external R&D strategies on thedistribution of profits. Main attributes of the productivity distribution are the dispersion, skewness andkurtosis. Quantile-based definitions of these attributes are as follows (see Oja, 1981 and Ruppert, 1987):

dispersion = (q0.75 − q0.25) / (q0.75 + q0.25)skewness = (q0.75 + q0.25 − 2q0.5) / (q0.75 − q0.25) (6)

kurtosis = (q0.90 − q0.10) / (q0.75 − q0.25)

The dispersion is a ratio of the width of the distribution between the upper and lower quartiles over a measureof location. The skewness compares the difference between the upper quartile and median and the medianand the lower quartile over the width of the distribution. This measure is zero for symmetric distributions.A negative value implies that the productivity distribution has longer tails on the left side but that the massof the distribution is concentrated on the right. The kurtosis measures the weight of the tails by comparingthe distance between the 0.10 and 0.90 quantiles with the distance between the upper and lower quartiles. Ahigh kurtosis points to a productivity distribution where the dispersion of productivity results from extremebut infrequent productivity levels (extreme deviations) whereas a low kurtosis implies that the dispersionresults from frequent modestly-sized deviations.

Inserting the equations for different quantiles into these definitions, we obtain a relationship between ourexplanatory variables and the distributional characteristics. In order to evaluate how changes in humancapital and innovation ceteris paribus affect these distributional characteristics, we follow Mata and Wörter(2013) by using the estimated coeffi cients of these variables at the relevant quantiles. Standard errors of thesenon-linear combinations of parameter estimates are calculated using the Delta method (Wooldridge, 2002).4

3 Data description

To ensure that our results reflect underlying economic differences, we built two highly comparable microdatasets that span the period 1998-2008. Enterprises in manufacturing (European industry classification systemNACE Rev. 1.1 15 to 37) and services (NACE 50 to 90) are included in the analysis. The population ofinterest consists of enterprises with at least ten employees. This section examines the German and Dutchmicrodata sets respectively. For both countries, price deflators for output, value added, intermediate inputs

4Calculations are done in STATA using the nlcom -command.

6

and capital are drawn from the EUKLEMS database (November 2009 release, March 2011 update) and unitlabor costs are taken from the OECD database.

3.1 Germany

We use the Mannheim Innovation Panel (MIP). The MIP is made up by representative innovation surveyswhich are collected by the Centre for European Economic Research (ZEW) in cooperation with the FraunhoferInstitute for Systems and Innovation Research (ISI) and the Institute for Applied Social Science (infas) onbehalf of the German Federal Ministry of Education and Research (BMBF). Every fourth (before 2005) /second (after 2005) year, the MIP is the German contribution to the European-wide harmonized CommunityInnovation Surveys (CIS). In contrast to other European countries, the MIP is an annual panel that startedin 1993 in manufacturing and was extended to services in 1995. It is based on a random stratified sample—industry, size and region serving as stratification criteria—that is refreshed every second year for dead andnewly established firms respectively (see Rammer and Peters, 2013). In addition to the common harmonizedinnovation indicators, the German innovation surveys additionally ask firms about a host of other general firmcharacteristics such as sales, number of employees, the share of high-skilled employees, intermediate inputcosts (including energy costs and intermediate services) and the stock of tangible assets (physical capital).

3.2 The Netherlands

We use data that are sourced from different surveys collected by Statistics Netherlands, or “Centraal Bureauvoor de Statistiek”(CBS). The innovation variables stem from five waves of the Dutch Community InnovationSurveys (CIS): CIS3 (1998-2000), CIS3.5 (2000-2002), CIS4 (2002-2004), CIS4.5 (2004-2006) and CIS5 (2006-2008). CIS enterprises are merged with data from the Production Surveys (PS).5 The latter contains dataon production value, factor inputs and factor costs.

The CIS and PS data are collected at the enterprise level. A combination of census and stratified randomsampling is used for each wave of the CIS and PS. A census is used for the population of enterprises with atleast fifty employees and a stratified random sampling is used for enterprises with fewer than fifty employees.The stratification variables are the industry and the number of employees of an enterprise. The same cut-offpoint of 50 employees is applied to each wave of the CIS and the PS.

The Social Statistics Database (SSB) forms the backbone to retrieve information on the skill composition ofthe workforce in the matched (CIS∩PS)-enterprises (Bakker, 2002). The SSB links administrative data forthe entire population registered as living in the Netherlands with detailed demographic and socio-economicdata from business and household surveys. The data are primarily obtained from the population register, taxregisters, social security registers, education registers and various other registers and administrations. TheSSB contains all the relevant information on persons, families, households, jobs, benefits and living quarterswhich can be matched with enterprise data through a unique personal identification number. Details on themeasurement of the human capital variables are found in Section A.1 of Appendix A.

3.3 Main estimation samples

For estimation purposes, we use information from the aforementioned five waves of the CIS (Germany)and matched CIS samples (The Netherlands) in both countries. After some cleaning and trimming on

5Approximately 26% of the CIS enterprises are matched with the corresponding PS enterprises in manufacturing. For services,the match increases to 33%.

7

nominal labor productivity levels and growth rates to eliminate outliers and anomalies, we end up withan unbalanced panel of 11,699 observations corresponding to 6,634 enterprises (61.4% in manufacturing and38.6% in services) over the period 2000-2008 in Germany (DE) and an unbalanced panel of 24,586 observationscorresponding to 14,841 enterprises (38.5% in manufacturing and 61.5% in services) over the period 2000-2008in the Netherlands (NL).6 The estimation samples are further broken down into five industries according tothe OECD (2001) classification: High-technology manufacturing (HT ), Medium-technology manufacturing(MT ), Low-technology manufacturing (LT ), Knowledge-intensive services (KIS ) and Other services (OS ).Table A.6 in Appendix A provides details on the industry breakdown of manufacturing and services dependingon their technological intensity.

Table 1 reports the number of observations and firms in the estimation sample by country, industry, size andyear. Unsurprisingly, the German sample includes more larger enterprises (10.9% with more than 500 employ-ees) than the Dutch sample (3.6%). With respect to industry composition, we find that the German sampleincludes more High-technology manufacturing firms but less Other services firms. That is, in DE (NL), 9.2%(2.7%) of the firms belong to High-technology manufacturing, 35.2% (22.6%) to Medium-technology manu-facturing, 16.6% (10.5%) to Low-technology manufacturing, 24.1% (29.7%) to Knowledge-intensive servicesand 14.9% (34.8%) to Other services. In some robustness checks and to measure some variables (see Section3.4), we use a more detailed industry classification (21 industries: 11 in manufacturing and 10 in services).Table B.1 in Appendix B presents the number of observations and the number of firms in the estimationsample by country and by 21-industry. Table B.2 in Appendix B gives the panel structure of the estimationsample. In DE (NL), 46% (38.2%) of the enterprises have at least two observations. For about 8% of theenterprises, we have at least four observations in the two countries.

<Insert Table 1 about here>

3.4 Dependent and explanatory variables

Our main dependent variable is the logarithm of real labor productivity (RLP ). Nominal labor productivity

is measured by sales per employee(QL

)where L is the number of employees in head counts.7 EUKLEMS

output price indicators (base year 2006) are used for deflation.

We explain the logarithm of real labor productivity by firm size (lnLit = SIZEit) and the traditional inputfactors physical capital and material. Capital is measured as the logarithm of real physical capital peremployee

(ln(KL

)it= CAPit

), where K is proxied by tangible assets in the German microdata set and by

depreciation of fixed assets in the Dutch microdata set. It is deflated by using the industry-level gross fixedcapital formation price index for all assets. Material is defined as the logarithm of real material costs peremployee

(ln(ML

)it=MATit

), where M is intermediate input costs including energy costs and intermediate

services, deflated by the industry-level intermediate inputs price index. In order to investigate the role ofhuman capital, we include the share of high-skilled labor (HCit), where high-skilled employees are defined ashaving a college or university degree. Innovation is captured by two innovation outcome variables: productand process innovation. Product innovation is measured by the logarithm of real innovative sales per employee(ln(SSPD×SALES

L

)it= PDit

). SSPDit refers to the share of total sales in year t accounted for by new or

6 In DE (NL), 2,506 (4,452) enterprises take part in at least two consecutive waves, 956 (1,860) in at least three consecutivewaves, 390 (785) in at least four consecutive waves and 152 (348) in all five waves.

7L refers to the average number of employees in the German data set and to the number of employees in September of agiven year in the Dutch data set.

8

improved products and services introduced in t, (t − 1) and (t − 2). In addition, we make a correspondingdistinction based on the share of sales due to products new to the firm only (firm novelties, SSFNit) andthe share of sales due to products new to the market (market novelties, SSMNit). In contrast to productinnovation, process innovation is measured by a binary indicator equaling one if an enterprise introducedany new or significantly improved production technology during the period under review, i.e. between (t− 2)and t (PCit). In order to investigate whether distance to the technological frontier matters for firm-levelproductivity, we include the the 1-year lagged value of closeness to the technological frontier (CTFit−1 =

L1.CTF ). Closeness to the technological frontier is measured as CTFit = 1−DTFit = 1−(RLPFt−RLPit

RLPFt

)=

RLPitRLPFt

, where RLP of the technological frontier firm F is proxied by the 95% percentile value of RLP at theNACE 3-digit industry level in both countries.8 The definition of L1.CTF implies that we capture persistenceeffects. Finally, our productivity estimates control for being part of a group (GPit), being located in EastGermany for DE (EASTit) and time dummies (Dt). In the estimations, our main focus is on the effect ofthe human capital and the innovation variables.

Despite the same definitions, one important difference between the German and Dutch variables stems fromthe measurement of the human capital variable. For DE, the skill variable is directly taken from the surveyinformation. For NL, this variable is mainly estimated using a matched employer-employee dataset (seeSection A.1 in Appendix A). In addition to this measurement issue, there are inherent institutional differencesbetween the two countries. In particular, the education system in DE is characterized by a dual system —integrating work-based and school-based learning—supportive for providing high-quality technical skills andfor creating a high degree of specialization of skilled employees.

In addition to the main model specification, we perform various robustness checks in Section 5.3. Thesensitivity analyses particularly refer to the measurement of the dependent variable and of human capital. Weexamine two alternative dependent variables. The first is total factor productivity (TFP ) which is calculatedas the residual of a panel estimation of a standard Cobb-Douglas production function at the industry level.We adopt the system generalized method of moments (SY S-GMM) estimator and use appropriate lags ofthe input factors as instruments. More specifically, we estimate a production function for each of the 35NACE 2-digit industries in DE and each of the 149 NACE 3-digit industries in NL and and calculate TFPas TFPit = ln (RLP )it − γ̂K ln

(KL

)it− γ̂M ln

(ML

)it− γ̂L lnLit −

∑tγ̂DDt.9 The second is the one-year lead

of real labor productivity growth (RLPGR), defined as labor productivity growth between year t and (t+1).

Regarding human capital, HCit is either replaced by (i) a binary variable equaling one if HCit exceeds themedian value of the share of high-skilled labor in industry j (21-industry classification) at time t or (ii) amore detailed decomposition of the workforce. This detailed decomposition is only feasible for NL and splitsL into the number of low-skilled, low-medium-skilled, high-medium-skilled and high-skilled employees.10

We furthermore investigate whether real labor productivity can be explained by different moments of theindustry distribution of human capital intensity (where industries are defined according to the 21-industry

8 In DE, we consider the largest possible population of enterprises included in the MIP. In addition to the response sample,this also includes information from the non-response sample. In total, 84, 454 observations from 19, 351 enterprises were usedfor calculating annual CTF during the period 1998-2008. For details on the measurement of CTF in NL, we refer to SectionA.2 of Appendix A.

9The number of observations for several 3-digit industries is insuffi cient to allow for estimations at a more detailed disaggre-gation level in DE.10Details on the definition of the four skill types are provided in Section A.1 of Appendix A.

9

classification). In particular, we consider the mean (HCmeanJt), the standard deviation (HCsdJt), theskewness (HCskewJt) and the kurtosis (HCkurtJt) of industry-year distributions of human capital intensity.

Table 2 shows descriptive statistics in the estimation samples for our key variables by country and by industry.Focusing on our dependent and primary explanatory variables, we observe considerable heterogeneity acrosscountries and —within a country—across industries. Except for Other services, average RLP is higher for allindustries in NL. In manufacturing, real labor productivity (both in levels and growth rates) varies much moreacross industries in NL, while the opposite is true for services. In both countries, average RLP decreaseswith the level of technological intensity in manufacturing. The same is true for services in DE whereasaverage RLP is the same in both service industries in NL. Over the period 2000-2008, real labor productivitygrows at an average annual rate of 3.6% in DE and 5.3% in NL. Except for Low-technology manufacturing,average RLPGR is significantly higher for all industries in NL. The relationship between average RLPGRand technological intensity appears to be hump-shaped in German manufacturing whilst average RLPGRincreases with the level of technological intensity in Dutch manufacturing. Average RLPGR is observed tobe higher in Knowledge-intensive services compared to Other services in DE whereas no difference can bedetected in NL.

The average share of high-skilled labor is 0.19 in DE and 0.26 in NL. A comparable difference in the averageproportion of individuals (aged 15-64) with tertiary educational attainment over the period 2000-2008 isreported by Eurostat (2013), i.e. 0.20 in DE and 0.24 in NL, which suggests that measurement differences inour human capital variable between the two countries (see supra) do not give any obvious cause for concerns.11

We observe considerable heterogeneity across industries. In both countries, high-technology enterprises inboth manufacturing and services possess a significantly higher fraction of high-skilled labor compared to theirlow-technology counterparts.

In contrast to human capital, we find that the proportion of innovators, either defined in terms of productinnovators or process innovators, and the share of innovative sales (SSPD) are on average higher in DE thanin NL. 64% and 42% of the enterprises in the German and Dutch sample, respectively, report having processor product innovation. In DE (NL), the proportion of innovators ranges from 38% (29%) in Other services to88% (66%) in High-technology manufacturing. The average share of sales due to products new to the market(SSMN) is slightly higher in NL, whereas the average share of sales due to products new to the firm only(SSFN) is much higher in DE. Comparing the different industries across countries reveals a clear pattern:the proportion of product and process innovators is higher in all German industries whilst the opposite istrue for the proportion of enterprises having introduced market novelties in Low-technology manufacturingand Other services. Focusing on innovation performance, the share of innovative sales is higher in all Germanindustries. Looking at the different types of product innovation, however, the numbers reveal that the shareof sales due to market novelties is considerably higher in all Dutch industries, suggesting that innovationsare more radical in NL.


11Corroborative evidence on NL outperforming DE in terms of skill levels based on international test scores is given in Minneet al. (2007).

10

4 Distributions of human capital intensity, innovation and produc-tivity: Some stylized facts

The productivity literature provides ample evidence that performance in terms of productivity is highlyskewed across firms and that this heterogeneity is persistent over time (see Bartelsman and Doms, 2000 for asurvey). This observation implies that persistent market dominance of firms is a pervasive fact in technologi-cally advanced countries (e.g. Clements and Ohashi, 2005). The ubiquity of firm-level productivity variationand persistence in itself has spurred research into the underlying factors shaping the firm productivity dis-tribution (see Syverson, 2011 for a survey). In this study, we are particularly interested in the role of humancapital and innovation in boosting productivity, both across countries and across industries.

This section presents some stylized facts on human capital intensity, innovation and productivity in bothcountries which serve as the backbone of the econometric analysis. More specifically, we provide a detailedcomparison of the distributions of human capital intensity, innovation and productivity across the two coun-tries and across industries. When discussing the moments of these distributions, we take the standard normaldistribution as the benchmark.

4.1 Human capital intensity distribution

Graph 1 presents the kernel density estimates of the distributions of human capital intensity by country andby industry. Table 3 reports the moments (mean, variance, skewness and kurtosis) of the corresponding dis-tributions.12 Focusing on cross-country differences, the average share of high-skilled employees is significantlyhigher in all Dutch industries. The difference varies between 4.6 and 10.1 percentage points in High- and Low-technology manufacturing respectively. This result is in line with OECD statistics on tertiary educationalattainment levels in both countries. During the period 2000-2006 about 23-24% of the German populationaged between 25-64 attained a tertiary degree. In NL, this proportion raised from 23.4 to 30.2% in the sameperiod (OECD, 2009). We observe considerably higher dispersion in all German industries, suggesting moreinequality in the distribution of human capital intensity in DE, as indicated by the coeffi cient of variation. InDE, the distribution of human capital intensity shows a right-skewed shape in all industries. In NL, we observethe same pattern, except for Knowledge-intensive services where the mass of the distribution of human capitalintensity is concentrated on the right. The positive skewness is significantly larger in all German industries.In both countries, the distribution of human capital intensity appears to be heavy-tailed in Medium- andLow-technology manufacturing and Other services, as indicated by the positive excess kurtosis.13 In thoseGerman industries, the positive excess kurtosis is much higher than in the Dutch counterparts, implying thatmore of the variance is due to extreme deviations. In line with expectations, High-technology manufacturingand Knowledge-intensive services are characterized by light-tailed distributions of human capital intensity.

Focusing on industry differences, the average share of high-skilled employees is the lowest in Low-technologymanufacturing and the highest in Knowledge-intensive services followed by High-technology manufacturing.The coeffi cient of variation, however, shows that human capital intensity is less dispersed in High-technologymanufacturing than in Knowledge-intensive services. The highest dispersion of human capital intensity

12When interpreting Graph 1 and Table 3, one should keep in mind that if all firms used human capital at the same intensity,the distribution of human capital intensity would degenerate at one mass point.13 In order to compare the distribution with a standard normal distribution which has a kurtosis (k) of k = 3, the excess

kurtosis (ke) is defined as ke = k − 3.

11

among firms is observed in Other services. This industry distribution pattern holds for both countries. Asalready mentioned, the human capital intensity distribution is right-skewed in all industries, except for theDutch Knowledge-intensive services. It is characterized by the highest positive skewness in Low-technologymanufacturing in both countries whereas the distribution of human capital intensity in Knowledge-intensiveservices shows a light right-skewed shape in DE and even a left-skewed shape in NL. The distribution of humancapital intensity is light-tailed in Knowledge-intensive services whilst most heavily tailed in Low-technologymanufacturing in both countries.

<Insert Graph 1 and Table 3 about here>

4.2 Innovation performance distribution

While human capital intensity is consistently higher in the NL compared to DE, we observe the oppositepattern with respect to innovation performance. Graph 2 presents the kernel density estimates of the inno-vation performance distributions for product innovators by country and by industry. Table 4 completes thispicture by reporting the related moments of the distributions. As mentioned above, innovation performanceis measured by the logarithm of real innovative sales per employee for product innovators.

Interesting cross-country and cross-industry differences show up. Innovation performance is on average higherand at the same time less dispersed in all German industries, except for Low-technology manufacturing. Themass of the distribution is concentrated on the right in all German industries (left-skewed). The sameholds for the Dutch counterparts, except for Medium-technology manufacturing. The left-skewness is morepronounced in German High-technology manufacturing and Other services and in Dutch Low-technologymanufacturing and Knowledge-intensive services. In contrast to the mean and dispersion, we find mixedresults with respect to the kurtosis. Compared to a standard normal distribution with a kurtosis of 3, theinnovation performance distribution is more peaked and has longer heavier tails in all German industries,except for Other services where we observe a platykurtic distribution. In NL, the distribution is likewisemore peaked compared to a standard normal distribution in Medium- and Low-technology manufacturingand Knowledge-intensive services. This peakedness is less pronounced in the Dutch Medium-technologymanufacturing but it is stronger than in DE in the latter two industries. In contrast, the excess kurtosis isnegative in the Dutch High-technology manufacturing, indicating a relatively flat distribution.

Focusing on industry differences, we observe the same industry ranking in terms of average innovation per-formance in both countries. That is, average innovation performance is the highest in High-technologymanufacturing and the lowest in Other services. At the same time, innovation performance is less dispersedin High-technology manufacturing and most widely dispersed in Other services in both countries. As alreadymentioned, the distribution is left-skewed in all German industries. The negative skewness is the highest inHigh-technology manufacturing and the lowest —almost symmetric—in Low-technology manufacturing. In NL,the innovation performance distribution is most skewed to the left in Knowledge-intensive services, followedby Low-technology manufacturing. In contrast, we observe a right-skewed shape in Medium-technology man-ufacturing. The kurtosis is the highest in High-technology manufacturing and the lowest in Other services inDE whilst the opposite holds in NL.


12

4.3 Labour productivity distribution

Table 5 reports the moments of the labor productivity distribution by country and by industry. The firstpart of the table presents the distributional characteristics for all enterprises, the second part distinguishesbetween high-skilled and low-skilled enterprises and the third part between enterprises with a high and lowinnovation performance.14 The corresponding labor productivity differences are visualized in Graph 3.

Focusing on cross-country differences, average labor productivity is higher in all manufacturing industriesin NL whilst the opposite is true for Other services. Labor productivity is less dispersed in all Dutchindustries, except for High-technology manufacturing. We do not observe a clear pattern with respect tothe skewness of the labor productivity distributions across countries. In Low-technology manufacturing andKnowledge-intensive services, the distribution is left-skewed in both countries and more pronounced so inDE. In High- and Medium-technology manufacturing, we likewise observe a left-skewed distribution in DEwhilst it is skewed to the right in NL. In contrast, we find a right-skewed distribution in Other services inboth countries, which is even more pronounced in NL. The figures further reveal that the labor productivitydistributions consistently have sharper peaks and heavier tails than a standard normal distribution in allindustries in both countries. Except for Medium-technology manufacturing, this positive excess kurtosis ishigher in all Dutch industries.

Focusing on industry differences, we observe the lowest average labor productivity in Knowledge-intensiveservices in DE and in Other services in NL. In contrast, the highest average labor productivity is recorded inOther services in DE and in Medium-technology manufacturing in NL. While we do not observe a unifiedranking of industries in terms of average productivity in both countries, we find one in terms of dispersion.The lowest dispersion is detected in High-technology manufacturing and the highest dispersion in Knowledge-intensive services in both countries. Both the coeffi cient of variation and the difference between the 0.90 and0.10 quantiles lead to this conclusion. The latter indicates, e.g., that the 10% most productive firms inHigh-technology manufacturing are at least about 3.8 (DE ) to 4.8 (NL) times more productive than the 10%least productive firms. The labor productivity distribution is most skewed to the left in High-technologymanufacturing in DE. Among the left-skewed (right-skewed) distributions in NL, we observe the highestnegative (positive) skewness in Low-technology (Medium-technology) manufacturing. The distribution isleptokurtic in all industries. The lowest positive excess kurtosis is detected in Knowledge-intensive servicesin both countries whilst the highest positive excess kurtosis is recorded in Medium-technology manufacturingin DE and High-technology manufacturing in NL.

How can these differences in labor productivity distributions across countries and industries be explained?As already pointed out, we are particularly interested in the role of human capital and innovation in shapingproductivity. We therefore also differentiate between low- versus high-skilled and low- versus high-innovativeenterprises.

Focusing on the first two moments of the labor productivity distribution in the low- and high-skilled groups,we confirm that average labor productivity is consistently higher in high-skilled enterprises, except for Otherservices in DE. Labor productivity is less dispersed in high-skilled enterprises in all German industries whilstthis does not hold for Medium-technology manufacturing and Other services in NL.

14Enterprises with a share of high-skilled employees above the median are defined as high-skilled enterprises. Likewise, productinnovators with real innovative sales per employee exceeding the median are defined as enterprises with a high innovationperformance.

13

Distinguishing enterprises on the basis of their innovation performance, average labor productivity is consis-tently higher in all high-innovative enterprises in both countries. This is accompanied by a lower dispersionin these enterprises in DE, except for Low-technology manufacturing where no difference in dispersion can bedetected. In NL, the pattern is more heterogenous. Labor productivity is less dispersed in high-innovativeenterprises in High-technology manufacturing and Knowledge-intensive services whilst the opposite is true inthe other three industries.


So far, we looked at productivity distributions from a static point of view. When we shift our focus to theevolution of productivity distributions, a particular interesting feature is firm-level persistence. To get a firstinsight into the persistent market dominance of firms, we examine firm-level persistence in the closeness tothe technological frontier in both countries. Table 6 reports transition probability rates of productivity acrossstates over the period 2000-2008. The states are defined as the bottom four quintiles, the 80th-95th percentileand the frontier which captures the upper 5% of the distribution. We distinguish two samples: the population(which draws from the yearly Production Surveys in NL) and the estimation sample. For the former, wepresent one-year transition probability rates from period t to period (t+1) and two-year transition probabilityrates from period t to period (t + 2). For the latter, we only report two-year transition probability ratessince the CIS waves are on a biannual basis in NL. Overall, Table 6 reveals a relatively strong persistencein productivity as we observe the highest values on the diagonal for each of the six states. Focusing onthe population, firm-level persistence appears to be slightly higher in DE. For example, comparing two-yeartransition probabilities, 61.3% of the frontier firms remain in their initial state in DE whilst this is only truefor 52.6% of the frontier firms in NL. The observed persistence differences seem to disappear when focusingon the estimation sample, however. In both countries, frontier firms are fairly persistent: about 57% remainin their initial state.15


Focusing on industry differences in the persistence of frontier firms and on yearly transition probabilityrates in the population, we observe the lowest persistence in High-technology manufacturing and the highestpersistence in Knowledge-intensive services in NL: 60% of frontier firms remain in their initial state inthe former and 75% in the latter.16 In DE, we likewise find frontier firms to be least persistent in High-technology manufacturing (72%) but —in contrast to NL—frontier firms are most persistent in Low-technologymanufacturing (81%), followed by Knowledge-intensive services (78%). We observe the same industry rankingin the estimation sample in DE, i.e. frontier firms appear to be least and most persistent in High- (53%)and Low-technology manufacturing (53%) respectively. In NL, this pattern changes. Frontier firms are leastpersistent in Knowledge-intensive services and most persistent in Other services in NL: 48% of frontier firmsremain in their initial state in the former and 68% in the latter.

Summing up, this section illustrates considerable heterogeneity in productivity across the two countries,between different industries but also between enterprises within an industry. In the following section, we useeconometric tools to investigate the role of human capital and innovation in shaping productivity distributions.

15 In Table 6, CTF is based on real labor productivity. Transition rates where CTF is based on total factor productivity arefairly similar (see Table B.3 in Appendix B).16Transition matrices for individual industries are not reported but available upon request.

14

5 Results

As a benchmark, we first present average returns to human capital and innovation in Section 5.1. Ourmain results are reported in Section 5.2, where we first examine firm-level heterogeneity in these returnsand than exploit this degree of firm-level heterogeneity in order to describe how differences in human capitaland innovation returns shape industry-specific productivity distributions. Section 5.3 presents the results ofvarious robustness checks. We conclude with a discussion of the main results in Section 5.4.

5.1 Average returns to human capital and innovation

As a benchmark, we estimate average returns to human capital and innovation using Eq. (1). Tables 7and 8 present OLS and FE results respectively. From Table 7, it follows that the average return to HC issignificantly positive in both countries. However, an increase in the share of high-skilled employees, e.g. by 10percentage points, raises productivity more strongly in NL (+4.9%) than in DE (+1.2%). Table 7 also revealssubstantial heterogeneity in average HC returns across industries. In DE, we observe significantly positiveaverage HC returns in Medium- and Low-technology manufacturing and Other services but surprisinglynot in High-technology manufacturing and Knowledge-intensive services. Average HC returns are likewisesignificantly positive in all Dutch industries, except for High-technology manufacturing. In both countries,the average HC return decreases with the level of technological intensity of an industry. Except for Low-technology manufacturing, average HC returns are much higher in all Dutch industries. However, Table 8shows that average HC returns become insignificant, except for Medium-technology manufacturing in NLwhen we account for unobserved firm-specific effects. A relatively low within-variation in the human capitalvariable might explain this finding.

<Insert Tables 7 and 8 about here>

The OLS estimates also point to significantly positive average returns to product innovation in all industriesin both countries. The returns of a 1% increase in the product innovation performance range from 1.7%

(Medium-technology manufacturing) to 7.7% (Other services) in DE and from 0.6% (Medium-technologymanufacturing) to 5.5% (Other services) in NL. Except for Knowledge-intensive services, average returnsto product innovation are higher in all German industries. Moreover, service enterprises yield on average ahigher return in both countries. These significantly positive average returns to product innovation survive inall industries when we additionally account for firm-specific effects, except for High-technology manufacturingin DE. They shrink, however, to a range of about 0.7% to 2.9% in both countries.

From Table 7, it follows that average returns to process innovation are significantly negative in both countriesand larger in absolute terms in services.17 When accounting for unobserved firm-specific effects, the negativeaverage returns to PC become generally smaller and they only survive in Other services in both countriesand in Medium- and Low-technology manufacturing in DE.

17Admittedly, identifying the effect of process innovation is more diffi cult in empirical analyses. This is more likely to bethe case in service industries since services are more often customized to specific demands and clearly structured productionprocesses are lacking in many cases. Moreover, many enterprises perform product and process innovation simultaneously. Butwhile the PD variable is continuous, our PC variable is a binary indicator and hence less informative than PD. These tworeasons may partly explain the finding of a negative PC return.

15

5.2 Firm-level heterogeneity in returns to human capital and innovation and itsimpact on industry productivity distributions

To what extent do firm-level returns to human capital and innovation vary at different points of the conditionalproductivity distribution and how does this affect the characteristics of industry productivity distributions?We answer these two questions by first neglecting firm-fixed effects and using pooled quantile regressions(Section 5.2.1). In Section 5.2.2, we additionally account for firm-fixed effects in the quantile regressions.

5.2.1 Not accounting for firm-fixed effects

A. Firm-level heterogeneity in returns to human capital and innovation

Table 9 reports the results of pooled simultaneous-quantile regressions (QR) for the 10th, 25th, 50th, 75th and90th percentiles of the productivity distribution.18 Graphs 4, 5 and 6 display the estimated coeffi cients forour variables of interest (HC, PD and PC) across all quantiles, together with the 95% confidence intervals.For comparison, the OLS estimates and their 95% confidence intervals are presented as dotted horizontallines. Clearly, OLS estimates —calculating ‘the average effect for the average enterprise’—do not accuratelydescribe the relationship between our main variables and productivity. Let us focus the discussion on ourthree main variables.

The upper part of Table 9 and Graph 4 reveal a heterogeneous pattern for the effect of human capital uponproductivity at different quantiles. We observe heterogeneous productivity effects within an industry, andalso discern cross-country and cross-industry differences. In DE, the estimates point to an inverted U-shapedinfluence of human capital along the conditional productivity distribution in High- and Low-technologymanufacturing and Other services. This is particularly intriguing for High-technology manufacturing wherewe did not detect any significant average return. This can be explained by the fact that the 10% least-performing enterprises experience a significantly negative return to HC whereas enterprises along the 40th

and 80th percentile of the distribution yield significantly positive returns. In Low-technology manufacturingand Other services, we observe a different pattern: enterprises yield positive but first increasing and thandecreasing returns along the full conditional distribution. In Medium-technology manufacturing, we observeincreasing marginal returns to human capital as we move from lower to upper quantiles. The coeffi cient forHC starts negative (but insignificant) at the bottom of the distribution and becomes significantly positivefrom the 40th percentile onwards. On the contrary, we surprisingly observe diminishing rates of returnsto human capital in Knowledge-intensive services. Productivity effects of human capital are significantlypositive up to the 40th percentile, become insignificantly positive between the 40th and the 70th percentilesand are significantly negative from the 70th percentile onwards.

In contrast to DE, we do not find any such non-linearities in human capital returns in NL. We observeincreasing rates of returns to human capital as we move up through the productivity distribution in Medium-and Low-technology manufacturing and Other services. In the former two industries, the increase tends tobe steep whilst it is modest in the latter. Consistent with the German results, highly diminishing rates ofreturns to human capital are found in Knowledge-intensive services. From the 80th percentile onwards, theestimated coeffi cient for human capital is significantly negative.

18We estimate pooled simultaneous-quantile regressions for θ ∈ {0.05, 0.10, 0.20, 0.25, 0.30, 0.40, 0.50, 0.60, 0.70, 0.75, 0.80, 0.90, 0.95}.Table 9 shows results for some selected quantiles.

16

In a nutshell, the best-performing enterprises enjoy the highest rates of return to human capital in Medium-technology manufacturing in DE and in Medium- and Low-technology manufacturing and Other services inNL. This finding provides micro-economic support for the positive complementarity effect between humancapital and proximity to the technological frontier as postulated by Vandenbussche et al. (2006). In sharpcontrast, the top firms in Knowledge-intensive services seem to have the lowest (even negative) human capitalreturns in both countries, suggesting a negative aforementioned complementarity effect.

<Insert Table 9 and Graph 4 about here>

The middle part of Table 9 and Graph 5 highlight non-linearities in the returns to product innovation alongthe conditional productivity distribution in the majority of industries in both countries. In DE, we observean increase in the rates of returns to product innovation as we move from the lower to the upper quantilesin Low-technology manufacturing. In all other industries, marginal returns to product innovation follow aU -shaped curve.

In NL, the U -shaped pattern is likewise observed in Low-technology manufacturing and both service indus-tries. In contrast, a hump-shaped relationship is found in High-technology manufacturing with positive butinsignificant returns below the 20th and above the 80th percentile. Product innovation returns appear tobe very stable in Medium-technology manufacturing, although the estimated coeffi cient is not significantlydifferent from zero from the 75th percentile onwards.

Summing up, the best-performing enterprises enjoy the highest rates of return to product innovation inall industries in DE and in all but High- and Medium-technology manufacturing in NL, suggesting strongpositive complementarity effects between product innovation and proximity to the frontier.

<Insert Graph 5 about here>

The main finding that follows from the lower part of Table 9 and Graph 6 is that the top firms in themajority of industries experience the most negative rates of returns to process innovation. This holds forLow-technology manufacturing and both service industries in DE and for all industries, except for High-technology manufacturing in NL. In addition, the results shed some light on the negative average returnsto process innovation reported in Section 5.1. In all manufacturing industries in DE and in Low-technologymanufacturing in NL, they are caused by (extreme) outliers whilst the productivity effect of investing inprocess innovation is insignificant for most enterprises along the productivity distribution.


B. Impact on industry productivity distribution

To gain insight into the importance of human capital and product and process innovation in shaping thecharacteristics of industry productivity distributions, we combine firm-level results from regressions at differ-ent quantiles of the productivity distribution to evaluate how changes in these three variables ceteris paribusaffect the moments of industry productivity distributions.

The impact of human capital and both types of innovation upon the 2nd through 4th moment of the industryproductivity distributions are reported in Table 10. In both countries, human capital is found to exert asignificantly positive effect on the dispersion and the kurtosis of the productivity distribution in Medium- andLow-technology manufacturing whilst it leaves the skewness unchanged. Put differently, strategies to invest

17

in human capital do not only increase the median return in these industries (see Table 9) but also widenthe productivity distribution. This increased dispersion results from more extreme productivity outcomesat both right and left tails. In DE, we identify the same qualitative impact of product innovation on theproductivity distribution in Medium-technology manufacturing and Other services. This means that (i)productivity is significantly more dispersed for firms in these industries that invest in product innovation and(ii) this increased variability results from an increased mass at both tails of the productivity distribution.The latter effect is much stronger for product innovation than for human capital. In NL, the same qualitativeimpact of process innovation on the productivity distribution —i.e. a positive influence on the dispersion andthe kurtosis—is found in Medium- and Low-technology manufacturing.19

In addition, product innovation positively affects all three moments of the productivity distribution in Low-technology manufacturing in DE and Other services in NL. The finding that product innovation additionallyalters the skewness of the distribution means that the increased dispersion results from more extreme pro-ductivity outcomes at both tails but that we observe an increase in the concentration of mass on the right.

On the contrary, human capital is found to exert a negative effect on the dispersion, skewness and kurtosisof the productivity distribution in Dutch Knowledge-intensive services.


5.2.2 Accounting for firm-fixed effects

A. Restriction of estimation sample

To additionally account for unobserved firm heterogeneity in estimating human capital and innovation returns,we perform FE quantile regressions. For that purpose, we restrict the sample and only select firms with atleast two observations. We end up with an unbalanced panel of 8,117 observations corresponding to 3,052enterprises (61% in manufacturing and 39% in services) over the period 2000-2008 in DE and an unbalancedpanel of 15,427 observations corresponding to 5,664 enterprises (42% in manufacturing and 58% in services)over the period 2000-2008 in NL.

To investigate the selectivity impact of this restricted estimation sample, we performed the same analysis asfor the main estimation sample. The results of the first part, examining the average and quantile productivityeffects of human capital and innovation at the firm level, are largely confirmed when moving to the restrictedsample (see Table B.4 in Appendix B for the OLS and the FE results and Table B.5 in Appendix B for thestandard QR estimates). Likewise, the results of the second part, evaluating the impact of human capitaland innovation on the distributional characteristics of industry productivity, are mostly confirmed (see TableB.6 in Appendix B). In particular, results corroborate that human capital exerts a positive effect on thedispersion and kurtosis in Low-technology manufacturing in both countries and in Dutch Medium-technologymanufacturing. In Germany, a similar pattern is also observed in Medium-technology manufacturing andOther services. Results for product innovation are fully confirmed in both countries. That is, productinnovation positively affects the dispersion and kurtosis in Other services in both countries and in GermanMedium- and Low-technology manufacturing. In contrast, we do not find any distributional impact of processinnovation anymore. This may be explained by the fact that the influence of process innovation was largelydriven by some extreme outliers which may be dropped from the restricted sample.

19Mata and Wörter (2013) report a similar pattern for the impact of external innovation strategies on profits.

18

B. Firm-level heterogeneity in returns to human capital and innovation

Table 11 reports the results of estimating FE quantile regressions for the 10th, 25th, 50th, 75th and 90th per-centiles of the productivity distribution. For the sake of parsimony, we only report the estimated coeffi cientsfor HC, L1.CTF , PD and PC. A visual representation is given in Graphs 7-9. For comparison, the standardFE estimates and their 95% confidence intervals are presented as dashed horizontal lines. Similar to theOLS estimates, it appears that standard FE estimates —making inferences about ‘the average enterprise’—mask important aspects of the relationship between human capital and innovativeness on the one hand andproductivity on the other hand.

The upper part of Table 11 and Graph 7 focus on the heterogeneous productivity effects of human capital. InDE, the shape of human capital returns in Medium- and Low-technology manufacturing and Other servicesis quite comparable to the one of the standard QR estimates. In sharp contrast, accounting for firm fixedeffects leads to strongly increasing rates of returns to human capital as we move up through the productivitydistribution in Knowledge-intensive services. The estimated coeffi cient is significantly positive from the 40th

percentile onwards.

Taking into account firm heterogeneity does not seem to affect the heterogeneous pattern in human capitalreturns in Medium-technology manufacturing and both service industries in NL. Similar to the standard QRresults, we observe an increase in human capital returns as we move from the lower to the upper quantilesin Low-technology manufacturing but —contrary to the standard QR results—human capital returns appearto be significantly negative at each quantile. Contrary to the standard QR results, we observe positive butdiminishing rates of returns to human capital in High-technology manufacturing.

Summing up, consistent with the standard QR results, we find evidence of a positive complementarity effectbetween human capital and proximity to the frontier in Medium-technology manufacturing in DE and inMedium- and Low-technology manufacturing and Other services in NL. Also consistent with the standardQR results, a negative complementarity effect between human capital and proximity to the frontier is found inKnowledge-intensive services in NL. In contrast to the standard QR results, the best-performing enterprisesin Knowledge-intensive services seem to enjoy the highest rates of return to human capital in DE.

<Insert Table 11 and Graph 7 about here>

From the middle part of Table 11 and Graph 8, it follows that taking into account firm heterogeneity does notaffect the shape of the returns to product innovation compared to the standard QR results in NL. In DE, wecorroborate a U -shaped influence along the quantiles of the productivity distribution in all industries, exceptfor Low-technology manufacturing. However, the U -shape has become wider and the returns to productinnovation have become very stable for a broader range of quantiles. Contrary to the standard QR results,we do not find an increasing effect of product innovation along the quantiles in German Low-technologymanufacturing anymore but a hump-shape when taking individual heterogeneity into account.

Consistent with the standard QR results, we observe non-linearities in the productivity effects of product in-novation in Low-technology manufacturing and both service industries and stable product innovation returnsin Medium-technology manufacturing in NL. In contrast to the standard QR results, we observe graduallyincreasing returns to product innovation in High-technology manufacturing in NL.

Summing up, consistent with the standard QR results, we confirm a positive complementarity effect betweenproduct innovation and proximity to the frontier in Medium-technology manufacturing and both serviceindustries in DE and in all industries, except for Medium-technology manufacturing in NL.

19


The lower part of Table 11 and Graph 9 show that taking into account firm heterogeneity mainly influencesthe shape of returns to process innovation in DE. We only observe significantly estimated coeffi cients inMedium-technology manufacturing and Other services. More specifically, we find negative decreasing ratesof returns to process innovation in Medium-technology manufacturing whilst an increasing shape is detectedin Other services. For the latter, the estimated returns are significantly negative up to the 60th percentile.

Consistent with the standard QR results, we observe decreasing rates of returns to process innovation inKnowledge-intensive services with the estimated coeffi cients being significantly negative from the 30th per-centile onwards in NL. Also consistent with the standard QR results, returns to process innovation follow ahump-shaped curve in Other services with significantly negatively estimated coeffi cients at each quantile inNL.

Summing up, we find negative complementarity effects between process innovation and proximity to thefrontier in Medium-technology manufacturing in DE and in both service industries in NL. The latter isconsistent with the standard QR results.


C. Impact on industry productivity distribution

Based on the FE quantile regression results, Table 12 reports the influence of human capital and productand process innovation upon various distributional characteristics. Taking into account unobserved firm het-erogeneity significantly alters the impact of our variables of interest in DE. We do not observe any longer thathuman capital increases the productivity dispersion and kurtosis in German Medium- and Low-technologymanufacturing. Instead, human capital even narrows the distribution in High-technology manufacturing. Wealready pointed out that the impact of product innovation has become very similar along different quantilesonce we account for individual heterogeneity. As a result, we do not find a significant influence on the charac-teristics of industry productivity distributions anymore, except for Other services. Both product and processinnovation affect the dispersion and the kurtosis in Other services negatively.

Similarly to DE, accounting for firm fixed effects changes the influence of our main variables on the distri-butional characteristics of productivity in NL. Consistent with the standard QR results, firms that investin human capital in Medium-technology manufacturing are characterized by a wider productivity dispersionthat is shaped by more infrequent productivity levels at both tails. We do not longer find a significantimpact of human capital on the dispersion and the kurtosis in Low-technology manufacturing and Otherservices. Instead, human capital is found to widen the industry productivity distribution in Other services.Contrary to the standard QR results, the positive influence of product innovation on the dispersion and thekurtosis in High-technology manufacturing has disappeared but is now observed in Other services. Contraryto the standard QR results, we do not find any significant impact of process innovation on the distributionalcharacteristics of productivity when taking into account individual heterogeneity.


20

5.3 Robustness checks

To check the robustness of our results using the main estimation sample, we performed a large number ofsensitivity checks.20 The first set of robustness checks relates to our explanatory variables. Employing thelogarithm of real labor productivity as the dependent variable, we examined in both countries the productivityeffects of (i) different types of product innovation (i.e. market novelties versus firm novelties), (ii) humancapital where human capital is measured by a binary variable equaling one ifHCit exceeds the median value ofthe share of high-skilled labor in industry j (21-industry classification), (iii) human capital when additionallycontrolling for different moments of industry-year distributions of human capital intensity (where industriesare defined according to the 21-industry classification). In addition, we replaced the human capital variableand firm size in NL by a more detailed decomposition of the workforce, splitting the number of employeesinto the number of low-skilled, low-medium-skilled, high-medium-skilled and high-skilled employees.

In DE, our main results show a U -shaped pattern for product innovation returns along the conditional pro-ductivity distribution in all industries, except for Low-technology manufacturing where increasing returnsare found. It turns out that these results are to a large extent driven by market novelties. In particular, wecorroborate the results for this type of product innovation in all industries, except for High-technology man-ufacturing where the returns are steadily increasing when we move up through the productivity distribution.With respect to firm novelties, we still find evidence of U -shaped returns in Medium-technology manufac-turing and Knowledge-intensive services. In contrast, decreasing returns to firm novelties are observed inHigh-technology manufacturing and Other services. In addition, the returns to market novelties are higherthan the ones to firm novelties at nearly all quantiles in all industries in DE. In all industries, our resultssupport evidence of positive complementarity effects between market novelties and proximity to the techno-logical frontier. For firm novelties, this complementary effect only holds for Medium- and Low-technologymanufacturing and Knowledge-intensive services. Firm-level heterogeneity in the returns to market noveltiesalso significantly changes the distributional characteristics of productivity. We find a larger productivity dis-persion in all industries —except for Other services—that primarily stems from more infrequent productivitylevels at both tails.

In NL, we find increasing returns to market novelties in High- and Medium-technology manufacturing andin Knowledge-intensive services. Non-linearities in the productivity effects of investing in products new tothe market are observed in Low-technology manufacturing and Other services. Consistent with DE, thebest-performing enterprises enjoy the highest rates of returns to market novelties in all industries. Exceptfor High-technology manufacturing, innovation of products new to the market appear to have a significantlypositive impact on the dispersion and the kurtosis of the productivity distribution in all industries. We detectincreasing returns to firm novelties in all industries, except for High-technology manufacturing. In the latter,the productivity effects of investing in products new to the firm follow a U -shaped pattern. In contrast tomarket novelties, we only find a positive complementarity effect between firm novelties and proximity to thetechnological frontier in Medium- and Low-technology manufacturing. Innovation of products new to thefirm seems to exert a positive influence on the dispersion and skewness of the productivity distribution in allindustries, except for High- and Medium-technology manufacturing.

In NL, the productivity effects of human capital are robust to the measurement of human capital and tothe inclusion of additional covariates. In DE, we likewise confirm our main results with one exception.When measuring human capital by a binary variable, we do not find decreasing returns to human capital in

20Details on the results of these checks are available upon request.

21

Knowledge-intensive services any longer. Firms that are characterized by a human capital intensity above theindustry-median consistently enjoy positive HC returns along all quantiles of the productivity distribution.These returns are increasing up to the median and then start to decrease.

The second set of robustness checks examines the sensitivity of our main results to using two alternativedependent variables in both countries: (i) total factor productivity and (ii) the one-year lead of real laborproductivity growth. In DE, we fully confirm the results for the returns to product and process innovationwhen using TFP as the dependent variable. For human capital, however, evidence is mixed. We findincreasing returns to human capital in High- and Medium-technology manufacturing but in none of the otherindustries. In NL, the results using TFP as the dependent variable are qualitatively similar to the mainresults.

In DE, the results for human capital are qualitatively confirmed in High- and Medium-technology manufac-turing and Other services but not for the other two industries when using the one-year lead of real laborproductivity growth as the dependent variable. Estimates for the returns to product innovation becomeinsignificant for most quantiles in all German industries. In NL, the productivity effects of human capital arequalitatively confirmed in Medium-technology manufacturing and Knowledge-intensive services. In contrast,human capital returns lose significance in Low-technology manufacturing and Other services and becomesignificantly negative from the 75th percentile onwards in High-technology manufacturing. In contrast to themain results for NL, returns to product innovation only appear to be significant (and negative) from the50th percentile onwards in Low-technology manufacturing and returns to process innovation are significantlypositive in the lower quantiles in all industries, except for Medium- and Low-technology manufacturing.

5.4 Discussion

Focusing on the productivity effects of human capital, two main findings stand out. First, we observeincreasing marginal human capital returns to the best-performing enterprises in industries with a low levelof technological intensity in both countries. Having high-skilled employees makes it easier for frontier firmsin these industries to excel. Put differently, we find a significantly positive complementarity between humancapital and proximity to the frontier in these industries. Second, we observe diminishing (even negative)marginal human capital returns for the “stars” in Knowledge-intensive services, suggesting a significantlynegative complementarity between human capital and proximity to the frontier.21 Becoming a “superstar”seems to be extremely diffi cult if one is already quite successful. Apparently, high-skilled employees arehired for much more than they are worth. We put forward two interpretations. Firstly, the results simplyreflect a misallocation of high-skilled employees. Secondly, winner-take-all behavior underlies this finding.Investment in intangibles in frontier firms of Knowledge-intensive services, using human capital intensely,might create a profitable breakthrough for one firm which could compensate the losses of many competitors.Suggestive evidence indicates that this interpretation might be more valid in NL as (i) more experimentationand radical innovation takes place in frontier firms in Knowledge-intensive services, (ii) the distribution ofhuman capital intensity is more left-skewed in Knowledge-intensive services and (iii) the sharply decreasinghuman capital returns appear to be even stronger in the high-technology Knowledge-intensive services towhich the telecommunication, computer and R&D services industries belong.

21This finding is less consistent across different estimates in DE, though (see FE quantile regression estimates and standardQR estimates using TFP as the dependent variable).

22

Focusing on the productivity effects of innovation, the main finding is that there are increasing marginalproduct innovation returns but negative marginal process innovation returns for the best-performing enter-prises in the majority of industries in both countries. Hence, the best strategy for frontier enterprises is tofocus on product rather than on process innovation. In addition, our results suggest that product innovationstrategies are risky in Other services in both countries implying that these strategies might lead to a largenumber of successful projects but also to a large number of unsuccessful ones in that particular industry. InDE, the same result holds for Medium- and Low-technology manufacturing.

6 Conclusion

This study reconsiders the relationship between human capital and innovation on the one hand and produc-tivity on the other hand. We examine firm-level heterogeneity in returns to human capital and product andprocess innovation across industries that differ in the level of technological intensity and across countries. Inaddition, we exploit the degree in this firm-level heterogeneity to evaluate the impact of human capital andproduct and process innovation upon the attributes of industry productivity distributions.

Irrespective of their level of technological intensity, industries in the Netherlands are characterized by alarger average proportion of employees possessing a college or university degree and industries in Germanyby a more unequal distribution of human capital intensity. Average innovation performance —measured bythe logarithm of real innovative sales per employee for product innovators— is higher in all industries inGermany, except for Low-technology manufacturing. The distribution of innovation performance appears tobe wider in the Netherlands. Average productivity turns out to be higher in all manufacturing industriesin the Netherlands. Productivity is more unequally distributed in all German industries, except for High-technology manufacturing.

In both countries, non-linearities in the productivity effects of investing in product innovation are found inthe majority of industries. Frontier firms enjoy the highest returns to product innovation in most industries.Investing in product innovation significantly increases the spread of the productivity distribution and theprobability of observations at both the right and left tails of the productivity distribution in Other services inboth countries as well as in Medium- and Low-technology manufacturing in Germany. In sharp contrast, themost negative returns to process innovation are observed in the best-performing enterprises of most industriesin both countries. Clearly, the best strategy for frontier firms is to focus on product rather than on processinnovation.

In Germany, non-linearities in the productivity effects of investing in human capital in High- and Low-technology manufacturing and in Other services stand up whilst human capital returns follow an increasinglinear curve as we move up through the productivity distribution in all industries in the Netherlands, exceptfor Knowledge-intensive services. A positive complementarity effect between human capital and proximityto the technological frontier is found in industries with a low level of technological intensity whilst a negativecomplementarity effect is observed in Knowledge-intensive services in both countries. The latter result does nolonger hold for Germany once individual unobserved heterogeneity is taken into account. Suggestive evidencefor the Netherlands point to a winner-takes-all interpretation of this finding. Productivity is significantlymore dispersed for enterprises that invest in human capital in Medium- and Low-technology manufacturing,which is caused by an increased mass of extreme (positive and negative) productivity outcomes in theseindustries.

23

References

[1] Acemoglu, D., P. Aghion and F. Zilibotti (2003), Vertical integration and distance to frontier, Journalof the European Economic Association, 1(2-3), 630-638.

[2] Acemoglu, D., P. Aghion and F. Zilibotti (2006), Distance to frontier, selection and economic growth,Journal of the European Economic Association, 4(1), 37-74.

[3] Aghion, P., N. Bloom, R. Blundell, R. Griffi th and P. Howitt (2005), Competition and innovation: Aninverted-U relationship, Quarterly Journal of Economics, 120(2), 701-728.

[4] Aghion, P., R. Blundell, R. Griffi th, P. Howitt and S. Prantl (2004), Entry and productivity growth:Evidence from micro-level panel data, Journal of the European Economic Association, 2(2-3), 265-276.

[5] Aghion, P., R. Burgess, S. Redding and F. Zilibotti (2008), The unequal effects of liberalization: Evidencefrom dismantling the license Raj in India, American Economic Review, 98(4), 1397-1412.

[6] Antenbrink, P., K. Burger, M. Cornet, M. Rensman and D. Webbink (2005), Nederlands onderwijs enonderzoek in internationaal perspectief, CPB Document 88.

[7] Arias, O., K. Hallock and W. Sosa-Escudero (2001), Individual heterogeneity in the returns to schooling:Instrumental variables quantile regression using twins data, Empirical Economics, 26(1), 7-40.

[8] Bakker, B.F.M. (2002), Statistics Netherlands’ approach to social statistics: The social statisticaldataset, Statistics Newsletter, 11, 4-6.

[9] Barro, R. and J.W. Lee (2010), A new data set of educational attainment in the world, 1950-2010, NBERWorking Paper 15902.

[10] Bartelsman, E.J. and M. Doms (2000), Understanding productivity: Lessons from longitudinal micro-data, Journal of Economic Literature, 38(3), 569-594.

[11] Benhabib J. and M. Spiegel (1994), The role of human capital in economic development: Evidence fromaggregate cross-country data, Journal of Monetary Economics, 34(2), 143-174.

[12] Buchinsky, M. (1994), Changes in the US wage structure 1963-1987: Application of quantile regression,Econometrica, 62(2), 405-458.

[13] Buchinsky, M. (2001), Quantile regression with sample selection: Estimating women’s return to educa-tion in the US, Empirical Economics, 26(1), 87-113.

[14] Bulut, H. and G. Moschini (2009), US universities’net returns from patenting and licensing: A quantileregression analysis, Economics of Innovation and New Technology, 18(2), 123-137.

[15] Chevalier, A., C. Harmon, I. Walker and Y. Zhu (2004), Does education raise productivity, or just reflectit?, Economic Journal, 114(499), F499-F517.

[16] Choi, K.S. and J. Jeong (2007), Does unmeasured ability explain the wage premium associated withtechnological change?: Quantile regression analysis, Applied Economics, 39(7), 1163-1171.

[17] Clements, M. and H. Ohashi (2005), Indirect network effects and the product cycle: Video games in theU.S., 1994-2002, Journal of Industrial Economics, 53(4), 515-542.

24

[18] Coad, A. and R. Rao (2006), Innovation and market value: A quantile regression analysis, EconomicsBulletin, 15(13), 1-10.

[19] Coad, A. and R. Rao (2008), Innovation and firm growth in high-tech sectors: A quantile regressionapproach, Research Policy, 37(4), 633-648.

[20] Cohen, D. and M. Soto (2007), Growth and human capital: Good data, good results, Journal of EconomicGrowth, 12(1), 51-76.

[21] Coulombe, S., J. F. Tremblay and S. Marchand (2004), Literacy scores, human capital and growth acrossfourteen OECD countries, Statistics Canada and Human Resources and Skills Development Canada,Ottawa.

[22] de la Fuente, A. (2011), Human capital and productivity, Barcelona Economics Working Paper 530.

[23] de la Fuente, A. and R. Doménech (2001), Schooling data, technological diffusion and the neoclassicalmodel, American Economic Review, 91(2), 323-327.

[24] de la Fuente, A. and R. Doménech (2006), Human capital in growth regressions: How much differencedoes data quality make?, Journal of the European Economic Association, 4(1), 1-36.

[25] Denny, K. and V. O’Sullivan (2007), Can education compensate for low ability? Evidence from Britishdata, Applied Economics Letters, 14 (7-9), 657-660.

[26] Ebersberger, B., O. Marsili, T. Reichstein and A. Salter (2010), Into thin air: Using a quantile regressionapproach to explore the relationship between R&D and innovation, International Review of AppliedEconomics, 24(1), 95-102.

[27] European Commission (2008), Jobs and growth in the EU, A roadmap to a sustainable tomorrow, Europeon the move.

[28] Eurostat (2013), Regional education statistics.

[29] Falk, M. (2012), Quantile estimates of the impact of R&D intensity on firm performance, Small BusinessEconomics, 39(1), 19-37.

[30] Flabbi, L., S. Paternostro and E. Tiongson (2008), Returns to education in the economic transition: Asystematic assessment using comparable data, Economics of Education Review, 27(6), 724-740.

[31] Fox, J. and V. Smeets (2011), Does input quality drive measured differences in firm productivity?,International Economic Review, 52(4), 961-989.

[32] Goedhuys, M. and L. Sleuwaegen (2009), High-growth entrepreneurial firms in Africa: A quantile re-gression approach, Small Business Economics, 34(1), 31-51.

[33] Griffi th, R., S. Redding and J. van Reenen (2003), R&D and absorptive capacity: Theory and empiricalevidence, Scandinavian Journal of Economics, 105(1), 99-118.

[34] Griffi th, R., S. Redding and J. van Reenen (2004), Mapping the two faces of R&D: Productivity growthin a panel of OECD industries, Review of Economics and Statistics, 86(4), 883-895.

25

[35] Hall, B., J. Mairesse and P. Mohnen (2010), Measuring the returns to R&D. In Handbook of the Eco-nomics of innovation, vol. 2, Hall B, Rosenberg N (eds). North-Holland: Amsterdam.

[36] Hall, B. (2011), Innovation and productivity, NBER Working Paper 17178.

[37] Hanushek, E. and D. Kimko (2000), Schooling, labor-force quality and the growth of nations, AmericanEconomic Review, 90(5), 1184-1208.

[38] Hanushek, E. and L. Wößmann (2008), The role of cognitive skills in economic development, Journal ofEconomic Literature, 46(3), 607-668.

[39] Hanushek, E. and L. Wößmann (2012), Do better schools lead to more growth? Cognitive skills, economicoutcomes and causation, Journal of Economic Growth, 17(4), 267-321.

[40] Harmon, C., H. Oosterbeek and I. Walker (2003), The returns to education: Microeconomics, Journalof Economic Surveys, 17(2), 115-155.

[41] Hartog, J., P.T. Pereira and J.A.C. Vieira (2001), Changing returns to education in Portugal during the1980s and early 1990s: OLS and quantile regression estimators, Applied Economics, 33(8), 1021-1137.

[42] Hölzl, W. (2009), Is the R&D behaviour of fast growing SMEs different? Evidence from CIS III data for16 countries, Small Business Economics, 33(1), 59-75.

[43] Inklaar, R., M. Timmer and B. van Ark (2008), Market services productivity across Europe and the US,Economic Policy, 23(53), 139-194.

[44] Kaiser, U. (2009), Patents and profit rates, Economics Letters, 104(2), 79-80.

[45] Kneller, R. and P.A. Stevens (2006), Frontier technology and absorptive capacity: Evidence from OECDmanufacturing industries, Oxford Bulletin of Economics and Statistics, 68(1), 1-21.

[46] Koenker, R.W. (2005), Quantile regression, Cambridge: Cambridge University Press.

[47] Koenker, R.W. and G.J. Bassett (1978), Regression quantiles, Econometrica, 46(1), 33-50.

[48] Kok, W. (2004), Facing the challenge: The Lisbon strategy for growth and employment, Report for ahigh-level group.

[49] Krueger, A. and M. Lindahl (2001), Education for growth: Why and for whom?, Journal of EconomicLiterature, 39(4), 1101-1136.

[50] Love, J.H, S. Roper and J. Dun (2009), Innovation, ownership and profitability, International Journalof Industrial Organization, 27(3), 424-434.

[51] Machado, J.A.F., and J. Mata (2000), Box-Cox quantile regression and the distribution of firm sizes,Journal of Applied Econometrics, 15(3), 253-274.

[52] Machado, J.A.F., and J. Mata (2001), Earning functions in Portugal 1982-1994: Evidence from quantileregressions, Empirical Economics, 26(1), 115-134.

[53] Machado, J.A.F., and J. Mata (2005), Counterfactual decomposition of changes in wage distributionsusing quantile regression, Journal of Applied Econometrics, 20(4), 445-465.

26

[54] Madsen, J., R. Islam and J. Ang (2010), Catching up to the technology frontier: The dichotomy betweeninnovation and imitation, Canadian Journal of Economics, 43(4), 1389-1411.

[55] Martins, P. and P. Pereira (2004), Does education reduce wage inequality? Quantile regression evidencefrom 16 countries, Labour Economics, 11(3), 355-371.

[56] Mason, G., B. O’Leary and M. Vecchi (2012), Certified and uncertified skills and productivity growthperformance: Cross-country evidence at industry level, Labour Economics, 19(3), 351-360.

[57] Mata, J. and M. Wörter (2013), Risky innovation: The impact of internal and external R&D strategiesupon the distribution of returns, Research Policy, 42(2), 495-501.

[58] Minne, B., M. Rensman, B. Vroomen and D. Webbink (2007), Excellence for productivity? CentraalPlanbureau, The Hague.

[59] Mwabu, G. and T.P. Schultz (1996), Education returns across quantiles of the wage function: Alternativeexplanations for returns to education by race in South Africa, American Economic Review, 86(2), 335-339.

[60] Nahm, J.W. (2001), Non-parametric quantile regression analysis of R&D-sales relationship for Koreanfirms, Empirical Economics, 26(1), 259-270.

[61] Nelson, R. and E. Phelps (1966), Investment in humans, technological diffusion and economic growth,American Economic Review, 56(1-2), 69-75.

[62] OECD (2001), Science, technology and industry outlook 2001. Paris: OECD.

[63] OECD (2009), OECD Factbook 2009, Paris: OECD.

[64] Oja, H. (1981), On location, scale, skewness and kurtosis of univariate distributions, ScandinavianJournal of Statistics, 8, 154-168.

[65] O’Mahony, M., C. Robinson and M. Vecchi (2008), The impact of ICT on the demand for skill labour:A cross-country comparison, Labour Economics, 15(6), 1435-1450.

[66] Pereira, P. and P.S. Martins (2002), Is there a return-risk relationship in education?, Economics Letters,75(2), 31-37.

[67] Rammer, C. and B. Peters (2013), Innovation Panel Surveys in Germany, in: Gault, F. (eds), Handbookof Innovation Indicators and Measurement, 135-177, Edward Elgar: Cheltenham, UK and Northampton,MA, USA.

[68] Ruppert, D. (1987), What is kurtosis? An influence function approach, The American Statistician, 41,1-5.

[69] Sapir, A., P. Aghion, G. Bertola, M. Hellwig, J. Pisani-Ferry, D. Rosati, J. Viñals and H. Wallace (2004),An agenda for a growing Europe: The Sapir report, Oxford: Oxford University Press.

[70] Segarra, A. (2010), Innovation and productivity in manufacturing and service firms in Catalonia: Aregional approach, Economics of Innovation and New Technology, 19(3), 233-258.

27

[71] Segarra, A. and M. Teruel (2011), Productivity and R&D sources: Evidence for Catalan firms, Economicsof Innovation and New Technology, 20(8), 727-748.

[72] Sianesi, B. and J. van Reenen (2003), The returns to education: Macroeconomics, Journal of EconomicSurveys, 17(2), 157-200.

[73] Stam, E. and K. Wennberg (2009), The roles of R&D in new firm growth, Small Business Economics,33(1), 77-89.

[74] Syverson, C. (2011), What determines productivity?, Journal of Economic Literature, 49(2), 326-365.

[75] Vandenbussche, J., P. Aghion and C. Meghir (2006), Growth, distance to frontier and composition ofhuman capital, Journal of Economic Growth, 11(2), 97-127.

[76] Wooldridge, J. (2002), Econometric analysis of cross sections and panel data, MIT Press: Cambridge,MA.

28

Table 1: Estimation sample by country, industry, size and year

GERMANY THE NETHERLANDSSample # obs. % # �rms % # obs. % # �rms %Industry High-technology manufacturing (HT) 1,063 9.1 609 9.2 716 2.9 409 2.7

Medium-technology manufacturing (MT) 4,213 36.0 2,336 35.2 6,091 24.8 3,362 22.6Low-technology manufacturing (LT) 1,910 16.3 1,103 16.6 2,665 10.8 1,555 10.5Knowledge-intensive services (KIS) 2,737 23.4 1,599 24.1 6,624 26.9 4,415 29.7Other services (OS) 1,776 15.2 987 14.9 8,490 34.5 5,167 34.8

Industrya) High-technology manufacturing (HT) 734 9.0 280 9.2 486 3.1 179 3.2Medium-technology manufacturing (MT) 2,977 36.7 1,100 36.0 4,275 27.7 1,543 27.2Low-technology manufacturing (LT) 1,287 15.9 480 15.7 1,753 11.4 635 11.2Knowledge-intensive services (KIS) 1,860 22.9 722 23.7 3,664 23.7 1,383 24.4Other services (OS) 1,259 15.5 470 15.4 5,249 34.1 1,924 34.0

Firm size 10-19 2,580 22.1 1,482 22.3 3,086 12.5 2,726 18.420-49 2,659 22.7 1,487 22.4 6,333 25.7 4,797 19.550-99 1,969 16.8 1,151 17.4 6,384 25.9 3,567 24.0100-249 2,075 17.7 1,170 17.6 5,910 24.0 2,482 16.7250-500 1,080 9.2 618 9.3 1,692 6.9 736 4.9500-999 652 5.6 360 5.4 726 2.9 314 2.11000+ 684 5.8 366 5.5 455 1.8 219 1.5

Year 2000 1,543 13.2 - 4,519 18.4 -2002 2,246 19.2 - 5,365 21.8 -2004 2,404 20.5 - 5,063 20.6 -2006 2,486 21.2 - 4,533 18.4 -2008 3,020 25.8 - 5,106 20.8 -

Total 11,699 100.0 6,634 100.0 24,586 100.0 14,841 100.0

Note: a) Sample constrained to �rms with at least 2 observations (DE: 3,052 �rms, 8,117 observations; NL: 5,664 �rms, 15,427 observations).

29

Table 2: Descriptive statistics by country and industry

GERMANY THE NETHERLANDSMANUFACTURING SERVICES TOTAL MANUFACTURING SERVICES TOTAL

Unit HT MT LT KIS OS HT MT LT KIS OS

RLP a) mill. e per emp. 0.160 0.176 0.182 0.125 0.200 0.167 0.178 0.224 0.249 0.139 0.139 0.173

RLPGR % 0.031 0.043 0.037 0.038 0.022 0.036 0.074 0.048 0.035 0.058 0.057 0.053

TFP -0.011 -0.018 -0.001 0.006 -0.018 -0.009 0.015 0.024 0.023 0.099 0.095 0.068

TFPGR % -0.527 0.082 -0.164 4.063 -0.641 0.831 -3.532 1.148 -1.300 0.125 1.184 0.486

(median) % (-0.059) (-0.041) (-0.010) (-0.043) (-0.036) (-0.037) (-0.193) (-0.200) (-0.200) (-0.155) (-0.110) (-0.152)

SIZEa) head counts 692.4 1098.5 199.5 256.7 569.0 637.5 224.8 126.6 128.4 243.7 134.4 163.9

(median) head counts (60.0) (94.0) (70.0) (31.0) (49.0) (61.0) (68.0) (64.0) (64.0) (70.0) (70.0) (68.0)

CAP a) mill. e per emp. 0.045 0.055 0.055 0.124 0.121 0.080 0.006 0.008 0.007 0.008 0.006 0.007

MAT a) mill. e per emp. 0.081 0.097 0.100 0.045 0.116 0.087 0.081 0.124 0.152 0.049 0.029 0.073

GP [0/1] 0.426 0.441 0.326 0.289 0.318 0.367 0.735 0.690 0.587 0.537 0.687 0.638

EAST [0/1] 0.365 0.314 0.329 0.407 0.365 0.351 - - - - - -

CTF b) [0-1] 0.406 0.435 0.401 0.366 0.358 0.399 0.471 0.501 0.483 0.377 0.381 0.428

CTFTFP c) [�1;1] -0.017 -0.036 -0.021 -0.025 -0.006 -0.025 0.012 0.025 0.016 0.093 0.071 0.058

HC [0-1] 0.295 0.129 0.088 0.387 0.092 0.192 0.392 0.186 0.121 0.437 0.210 0.261

Innovationd) [0/1] 0.883 0.733 0.608 0.588 0.385 0.640 0.663 0.607 0.509 0.359 0.294 0.423

Product innovationd) [0/1] 0.772 0.571 0.418 0.422 0.220 0.476 0.570 0.505 0.407 0.281 0.209 0.334

SSPDe) [0-1] 0.373 0.265 0.245 0.340 0.190 0.288 0.291 0.247 0.209 0.244 0.212 0.236

Market noveltiesd) [0/1] 0.499 0.332 0.193 0.209 0.076 0.257 0.394 0.335 0.266 0.173 0.124 0.213

SSMNe) [0-1] 0.117 0.075 0.059 0.097 0.042 0.081 0.125 0.114 0.082 0.122 0.097 0.108

Firm noveltiesd) [0/1] 0.675 0.480 0.361 0.361 0.180 0.405 0.351 0.289 0.246 0.172 0.128 0.199

SSFNe) [0-1] 0.256 0.190 0.187 0.242 0.147 0.207 0.136 0.103 0.100 0.108 0.101 0.105

PC [0/1] 0.489 0.451 0.371 0.338 0.231 0.382 0.405 0.386 0.364 0.205 0.176 0.263

Notes: Number of observations: 11,699 in DE and 24,586 in NL, except for TFP (10,921 in DE and 24,578 in NL) since TFP was estimated only for �rms with at leasttwo observations. All monetary values are in million Euro and in constant prices (base year 2006). a) Absolute (mean) values are reported. In the estimations, however, weuse logarithmic values in order to account for the skewness of the distribution. b) CTF based on RLP . c) CTF based on TFP . d) Innovation, product innovation, marketnovelties and �rm novelties are binary indicators. e) SSPD is the share of sales in year t due to new products introduced in t, (t� 1) and (t� 2) for product innovators. Inthe estimations, however, we use the logarithm of real innovative sales per employee for product innovators

�PD = ln

�SSPD�SALES

L

��as an explanatory variable. Analogue

for market and �rm novelties.

30

Table 3: Distribution of human capital intensity 2000-2008, by country and industry

GERMANY THE NETHERLANDSHC HT MT LT KIS OS TOTAL HT MT LT KIS OS TOTALMean 0.328 0.157 0.101 0.425 0.119 0.216 0.429 0.214 0.147 0.481 0.182 0.290Sd 0.186 0.129 0.109 0.306 0.154 0.220 0.182 0.155 0.093 0.282 0.154 0.240Skewness 0.545 2.045 2.836 0.153 2.291 1.518 0.142 1.234 0.974 -0.409 1.156 0.854Kurtosis 3.225 9.226 15.224 1.647 8.270 4.670 2.653 4.802 5.841 1.861 4.192 2.641CV 0.567 0.819 1.075 0.720 1.293 1.018 0.425 0.726 0.633 0.587 0.850 0.828p10 0.100 0.040 0.010 0.030 0.010 0.020 0.198 0.052 0.034 0.0001 0.022 0.038p25 0.190 0.080 0.030 0.140 0.028 0.060 0.289 0.103 0.081 0.230 0.060 0.105p50 0.320 0.120 0.071 0.410 0.060 0.133 0.421 0.175 0.135 0.558 0.140 0.210p75 0.430 0.200 0.130 0.700 0.150 0.300 0.570 0.296 0.197 0.714 0.268 0.436p90 0.600 0.320 0.220 0.880 0.313 0.560 0.648 0.419 0.267 0.800 0.396 0.689

Note: Human capital is measured as the share of high-skilled employees, de�ned as employees having a college oruniversity degree.

Table 4: Innovation performance distribution 2000-2008, by country and industry

GERMANY THE NETHERLANDSPD HT MT LT KIS OS TOTAL HT MT LT KIS OS TOTALMean 3.976 3.803 3.181 3.166 2.668 3.578 3.757 3.755 3.412 2.807 2.522 3.394Sd 0.970 1.168 1.175 1.296 1.531 1.231 1.232 1.334 1.164 1.451 1.571 1.416Skewness -0.529 -0.421 -0.038 -0.266 -0.257 -0.444 -0.307 0.182 -0.458 -0.573 -0.107 -0.264Kurtosis 3.978 3.338 3.188 3.314 2.420 3.356 2.797 3.251 3.308 3.457 3.559 3.732CV 0.244 0.307 0.369 0.409 0.574 0.344 0.328 0.355 0.341 0.517 0.623 0.417p10 2.742 2.190 1.684 1.471 0.136 1.975 2.041 2.075 1.832 0.962 0.720 1.615p25 3.443 3.119 2.266 2.495 1.522 2.856 2.972 2.925 2.714 1.966 1.602 2.570p50 4.047 3.818 3.197 3.239 2.961 3.668 3.865 3.681 3.540 2.923 2.466 3.455p75 4.571 4.566 3.972 4.021 3.727 4.399 4.608 4.569 4.159 3.808 3.522 4.287p90 5.147 5.298 4.561 4.849 4.489 5.054 5.334 5.593 4.751 4.619 4.483 5.056

Note: Innovation performance is measured as ln(real innovative sales per employee) for product innovators where realinnovative sales are measured in thousand e (in constant prices of 2005).

31

Table 5: Labor productivity distribution 2000-2008, by country and industry

GERMANY THE NETHERLANDSSample ln(RLP) HT MT LT KIS OS TOTAL HT MT LT KIS OS TOTALTotal Mean 5.053 5.126 4.871 4.724 5.184 4.985 5.120 5.407 5.225 4.725 4.630 5.050

Sd 0.587 0.712 0.877 1.135 1.104 0.881 0.630 0.710 0.797 0.975 0.794 0.883Skewness -0.451 -0.054 -0.330 -0.334 0.940 -0.259 0.106 1.114 -0.123 -0.014 1.008 0.072Kurtosis 5.904 6.923 4.893 4.694 4.946 6.333 6.767 6.615 6.140 4.986 6.146 5.381CV 0.116 0.139 0.180 0.240 0.213 0.177 0.123 0.131 0.152 0.206 0.171 0.175p10 4.300 4.293 3.801 3.306 3.999 4.007 4.353 4.682 4.405 3.642 3.814 4.134p25 4.754 4.683 4.465 4.200 4.427 4.515 4.750 4.961 4.713 4.321 4.185 4.568p50 5.111 5.140 4.889 4.663 5.062 5.015 5.096 5.285 5.153 4.704 4.534 5.006p75 5.378 5.531 5.317 5.318 5.714 5.452 5.461 5.733 5.683 5.136 4.959 5.489p90 5.656 5.994 5.962 6.120 6.400 5.991 5.925 6.295 6.265 5.843 5.529 6.119

High HCb) Mean 5.123 5.227 5.017 4.784 5.111 5.097 5.252 5.556 5.370 4.865 4.906 5.195Sd 0.529 0.702 0.844 0.729 0.999 0.750 0.570 0.772 0.775 0.745 0.872 0.886Skewness -0.474 -0.216 -0.199 0.040 0.192 -0.116 0.388 1.093 -0.182 1.440 1.251 0.636Kurtosis 7.602 6.883 5.173 8.980 4.545 6.356 4.198 6.104 7.520 6.767 5.138 4.855CV 0.103 0.134 0.168 0.152 0.195 0.147 0.108 0.139 0.144 0.153 0.178 0.171p10 4.484 4.388 4.091 4.031 3.922 4.236 4.566 4.787 4.592 4.174 4.025 4.275p25 4.953 4.808 4.587 4.368 4.357 4.622 4.918 5.043 4.905 4.452 4.360 4.621p50 5.139 5.236 4.983 4.663 5.069 5.100 5.220 5.414 5.284 4.719 4.730 5.064p75 5.411 5.677 5.410 5.117 5.801 5.516 5.532 5.938 5.795 5.086 5.233 5.641p90 5.608 6.069 6.166 5.669 6.179 5.993 5.950 6.494 6.361 5.781 6.017 6.334

Low HC Mean 4.892 5.000 4.664 4.644 5.264 7.415 4.907 5.201 4.938 4.549 4.452 4.886Sd 0.678 0.703 0.881 1.518 1.205 -0.143 0.664 0.551 0.762 1.181 0.683 0.849Skewness -0.161 0.146 -0.470 -0.213 1.343 -0.143 0.118 0.461 -0.038 -0.204 0.410 -0.697Kurtosis 4.135 7.693 4.609 2.744 4.599 5.740 9.603 5.022 4.919 3.374 5.938 5.331CV 0.139 0.141 0.189 0.327 0.229 0.208 0.135 0.106 0.154 0.260 0.153 0.174p10 4.089 4.191 3.564 2.737 4.262 3.697 4.146 4.562 3.996 2.871 3.728 3.907p25 4.447 4.592 4.159 3.479 4.448 4.353 4.524 4.867 4.522 3.913 4.105 4.493p50 4.908 4.993 4.743 4.662 5.003 4.850 4.899 5.153 4.884 4.692 4.435 4.955p75 5.278 5.381 5.177 5.818 5.611 5.350 5.242 5.493 5.402 5.221 4.789 5.383p90 5.739 5.822 5.641 6.603 7.001 5.953 5.662 5.920 5.949 5.919 5.236 5.833

High PDc) Mean 5.256 5.440 5.238 5.051 5.325 5.223 5.410 5.703 5.620 5.241 5.178 5.545Sd 0.409 0.525 0.679 0.750 0.707 1.148 0.574 0.745 0.641 0.753 0.832 0.718Skewness 0.639 0.642 0.783 0.832 1.453 1.266 0.569 1.200 0.394 0.944 1.082 0.985Kurtosis 6.019 6.020 3.808 3.851 7.774 4.766 3.010 5.261 2.765 4.371 4.140 4.872CV 0.078 0.097 0.130 0.149 0.133 0.220 0.106 0.131 0.114 0.144 0.161 0.129p10 4.807 4.820 4.566 4.265 4.657 4.034 4.750 4.941 4.813 4.464 4.357 4.772p25 5.032 5.100 4.772 4.472 4.766 4.427 4.939 5.163 5.102 4.716 4.585 5.043p50 5.178 5.386 5.135 4.975 5.300 5.007 5.301 5.528 5.542 5.082 5.003 5.425p75 5.466 5.787 5.571 5.343 5.801 5.636 5.853 6.160 6.084 5.548 5.605 5.950p90 5.724 6.117 6.234 6.177 6.062 6.732 6.239 6.632 6.482 6.175 6.249 6.463

Low PD Mean 4.874 4.888 4.678 4.370 4.704 5.099 4.873 5.218 5.107 4.511 4.450 4.859Sd 0.507 0.579 0.609 0.739 0.638 1.023 0.556 0.513 0.547 0.799 0.700 0.711Skewness -0.708 -0.086 -0.244 -0.181 0.414 0.798 0.204 0.605 -0.411 -0.935 0.693 -0.627Kurtosis 3.968 3.958 4.637 3.594 3.575 4.403 3.959 8.493 7.009 5.381 6.313 6.530CV 0.104 0.119 0.130 0.169 0.136 0.201 0.114 0.098 0.107 0.177 0.157 0.146p10 4.209 4.145 3.801 3.297 3.922 3.891 4.143 4.657 4.529 3.350 4.074 4.132p25 4.578 4.513 4.425 3.934 4.309 4.386 4.562 4.903 4.777 4.300 4.481 4.540p50 4.980 4.918 4.680 4.488 4.586 5.058 4.855 5.197 5.077 3.665 4.776 4.890p75 5.264 5.245 5.053 4.743 5.007 5.571 5.167 5.515 5.423 4.900 5.252 5.244p90 5.299 5.584 5.390 5.122 5.608 6.437 5.578 5.725 5.735 5.228 5.589 5.622

Notes: a) Labor productivity is measured as ln(real turnover per employee) where real turnover is measured in thousand e (inconstant prices of 2005). b) High HC: high-skilled enterprises, de�ned as enterprises with a share of high-skilled employees abovethe median. c) High PD: high-innovative enterprises, de�ned as product innovators with real innovative sales per employeeexceeding the median.

32

Table 6: Firm-level persistence in the closeness to the technological frontier, based on RLP (transition rates)

GERMANY THE NETHERLANDS<q20 q20 - q40 - q60 - q80 - ≥q95 <q20 q20 - q40 - q60 - q80 - ≥q95

<q40 <q60 <q80 <q95 <q40 <q60 <q80 <q95Populationa)

CTFt CTFt+1<q20 79.40 15.08 3.50 1.25 0.60 0.17 77.02 14.24 4.98 2.44 1.01 0.32q20 - <q40 14.83 62.74 16.87 4.23 1.13 0.19 13.81 61.76 16.46 5.39 2.18 0.39q40 - <q60 2.85 18.11 59.94 16.04 2.64 0.41 3.27 17.57 58.24 16.98 3.42 0.52q60 - <q80 1.11 4.04 16.98 63.87 12.98 1.01 1.75 4.91 17.51 60.64 14.04 1.15q80 - <q95 0.64 1.65 3.94 19.45 67.36 6.96 0.96 2.23 5.23 19.28 66.15 6.15≥q95 (frontier) 0.33 0.88 1.25 4.05 16.76 76.72 0.80 1.24 2.23 4.44 20.18 71.11CTFt CTFt+2<q20 70.14 20.21 5.84 2.53 0.97 0.31 61.96 20.27 9.69 5.00 2.20 0.88q20 - <q40 19.28 49.58 22.15 6.78 1.80 0.41 19.96 43.15 22.02 9.35 4.59 0.93q40 - <q60 4.62 22.46 46.60 21.35 4.31 0.65 6.70 24.73 39.76 21.64 6.05 1.12q60 - <q80 2.20 7.04 21.59 49.50 17.88 1.79 3.44 9.29 24.26 43.39 17.60 2.02q80 - <q95 1.18 2.69 6.61 26.21 52.82 10.48 2.18 4.27 8.84 26.91 49.18 8.63≥q95 (frontier) 0.74 1.67 3.44 6.59 26.28 61.28 2.31 2.59 4.12 8.77 29.65 52.56

Estimation sampleb)

CTFt CTFt+2<q20 64.62 24.34 6.52 2.89 1.38 0.25 67.27 18.93 8.16 2.96 1.88 0.79q20 - <q40 21.34 44.63 22.80 8.66 2.20 0.37 20.95 45.93 20.46 8.35 3.76 0.56q40 - <q60 5.49 23.27 43.20 21.84 5.73 0.48 5.19 23.24 43.20 22.08 5.47 0.82q60 - <q80 2.24 7.73 22.82 45.26 19.95 2.00 2.57 9.16 24.84 44.79 16.75 1.88q80 - <q95 1.53 3.05 6.10 26.78 51.36 11.19 1.35 3.60 8.46 29.34 47.97 9.27≥q95 (frontier) 0.96 1.92 4.33 5.77 30.29 56.73 2.97 2.04 2.60 9.11 25.46 57.81

Notes: CTF is divided into six categories based on the annual 20th, 40th, 60th, 80th and 95th percentiles of real labor productivity. a) DE: 61,741 observations, NL:269,092 observations. b) DE: 11,699 observations, NL: 24,634 observations.

33

Table 7: Mean regression (OLS): Firm-level returns to human capital and innovation, by country and industry


HT MT LT KIS OS HT MT LT KIS OSHC 0.006 0.149*** 0.423*** 0.028 0.241** 0.124*** 0.017 0.349*** 0.417*** 0.259*** 0.677*** 0.485***

(0.055) (0.057) (0.093) (0.039) (0.105) (0.025) (0.067) (0.036) (0.105) (0.028) (0.035) (0.023)L1.CTF 1.036*** 0.888*** 1.009*** 1.518*** 1.415*** 1.342*** 0.652*** 0.414*** 0.344*** 1.307*** 1.316*** 1.139***

(0.088) (0.048) (0.083) (0.070) (0.069) (0.030) (0.077) (0.032) (0.044) (0.043) (0.034) (0.020)PD 0.019*** 0.017*** 0.028*** 0.037*** 0.077*** 0.029*** 0.015*** 0.006*** 0.019*** 0.044*** 0.055*** 0.039***

(0.005) (0.002) (0.005) (0.004) (0.007) (0.002) (0.005) (0.002) (0.004) (0.003) (0.003) (0.001)PC -0.048** -0.026** -0.049** -0.087*** -0.100*** -0.043*** -0.049* -0.036* -0.047*** -0.115*** -0.096*** -0.089***

(0.022) (0.013) (0.022) (0.023) (0.033) (0.009) (0.027) (0.008) (0.018) (0.017) (0.014) (0.007)SIZE 0.034*** 0.012** -0.011 -0.083*** -0.037*** -0.013*** 0.001 -0.010** -0.003 -0.055*** -0.038*** -0.039***

(0.010) (0.005) (0.010) (0.010) (0.011) (0.004) (0.013) (0.005) (0.011) (0.006) (0.006) (0.003)CAP 0.045*** 0.055*** 0.061*** 0.085*** 0.010 0.055*** 0.118*** 0.100*** 0.114*** 0.158*** 0.091*** 0.113***

(0.009) (0.007) (0.010) (0.007) (0.009) (0.003) (0.018) (0.006) (0.013) (0.007) (0.006) (0.004)MAT 0.290*** 0.318*** 0.354*** 0.206*** 0.237*** 0.231*** 0.302*** 0.442*** 0.503*** 0.144*** 0.123*** 0.185***

(0.026) (0.017) (0.024) (0.010) (0.014) (0.007) (0.028) (0.014) (0.017) (0.005) (0.004) (0.003)GP 0.017 0.083*** 0.097*** 0.116*** 0.061* 0.083*** 0.082*** 0.046*** 0.068*** 0.107*** 0.100*** 0.094***

(0.026) (0.014) (0.027) (0.028) (0.033) (0.010) (0.029) (0.009) (0.016) (0.014) (0.010) (0.006)EAST -0.075*** -0.140*** -0.128*** -0.127*** -0.117*** -0.112*** - - - - - -

(0.024) (0.014) (0.025) (0.027) (0.032) (0.010)Const -1.510*** -1.274*** -0.933*** -0.975*** -0.859*** -1.304*** -0.811*** -0.314*** -0.007 -0.820*** -1.095*** -1.095***

(0.145) (0.078) (0.098) (0.084) (0.092) (0.046) (0.168) (0.064) (0.090) (0.063) (0.059) (0.059)Time dummies yes yes yes yes yes yes yes yes yes yes yes yesIndustry dummies no no no no no yes no no no no no yesR2 0.786 0.789 0.796 0.729 0.687 0.787 0.76 0.825 0.848 0.753 0.688 0.764RMSE 0.304 0.299 0.369 0.484 0.469 0.375 0.303 0.261 0.329 0.449 0.374 0.389# obs. 1,063 4,213 1,910 2,737 1,776 11,699 716 6,091 2,665 6,624 8,490 24,586

Notes: The dependent variable is ln(real labor productivity). Significance level of ***1%, **5%, *10%. Standard errors are heteroskedasticity consistent and clusteredby enterprises.

34

Table 8: Mean regression (FE): Firm-level returns to human capital and innovation, by country and industry


HT MT LT KIS OS HT MT LT KIS OSHC -0.082 -0.032 0.160 0.024 0.189 0.041 0.202 0.174*** -0.217 -0.022 0.038 0.053

(0.121) (0.085) (0.135) (0.062) (0.199) (0.042) (0.181) (0.057) (0.153) (0.049) (0.068) (0.035)L1.CTF 0.537*** 0.497*** 0.256*** 0.884*** 0.680*** 0.567*** 0.190** 0.214*** 0.223*** 0.623*** 0.545*** 0.443***

(0.116) (0.052) (0.057) (0.144) (0.121) (0.046) (0.082) (0.036) (0.081) (0.064) (0.054) (0.026)PD 0.007 0.017*** 0.011*** 0.029*** 0.023*** 0.020*** 0.015** 0.008*** 0.018*** 0.023*** 0.028*** 0.022***

(0.007) (0.003) (0.004) (0.006) (0.007) (0.002) (0.007) (0.002) (0.004) (0.003) (0.003) (0.002)PC -0.010 -0.036*** -0.007 -0.012 -0.043* -0.023*** -0.032 -0.002 -0.001 -0.018 -0.049*** -0.031***

(0.024) (0.012) (0.015) (0.020) (0.026) (0.009) (0.037) (0.008) (0.017) (0.014) (0.012) (0.007)SIZE -0.048 -0.148*** -0.343*** -0.216*** -0.264*** -0.184*** -0.135*** -0.165*** -0.251*** -0.336*** -0.228*** -0.294***

(0.071) (0.030) (0.050) (0.043) (0.049) (0.021) (0.039) (0.021) (0.054) (0.027) (0.023) (0.014)CAP 0.054*** 0.028*** 0.033* 0.009 0.018* 0.020*** 0.036 0.053*** 0.049*** 0.100*** 0.075*** 0.084***

(0.017) (0.011) (0.018) (0.006) (0.011) (0.004) (0.031) (0.010) (0.013) (0.011) (0.010) (0.006)MAT 0.208*** 0.156*** 0.116*** 0.053*** 0.061*** 0.087*** 0.358*** 0.447*** 0.423*** 0.117*** 0.104*** 0.147***

(0.068) (0.021) (0.030) (0.012) (0.014) (0.008) (0.040) (0.025) (0.061) (0.008) (0.006) (0.005)GP 0.005 0.037* 0.010 -0.017 0.043 0.017 -0.026 0.003 0.058*** 0.008 0.010 0.007

(0.059) (0.022) (0.038) (0.033) (0.039) (0.015) (0.035) (0.010) (0.019) (0.013) (0.011) (0.007)EAST -0.672*** -0.101 -0.345*** 0.291** - -0.073 - - - - - -

(0.053) (0.106) (0.053) (0.118) (0.086)Time dummies yes yes yes yes yes yes yes yes yes yes yes yesIndustry dummies no no no no no yes no no no no no yesρ 0.855 0.894 0.962 0.916 0.920 0.916 0.818 0.823 0.872 0.888 0.836 0.876R2within 0.347 0.378 0.288 0.324 0.332 0.284 0.670 0.715 0.545 0.677 0.418 0.525

Notes: The dependent variable is ln(real labor productivity). Significance level of ***1%, **5%, *10%. Standard errors are heteroskedasticity consistent. ρ denotes thefraction of the overall variance that is due to individual heterogeneity.

35

Table 9: Quantile regression (QR): Firm-level returns to human capital and innovation, by country and industry

GERMANY THE NETHERLANDSHT MT LT KIS OS TOTAL HT MT LT KIS OS TOTAL

HC q10 -0.181** -0.117 0.087 0.318*** 0.126* 0.018 -0.014 0.152*** 0.131* 0.582*** 0.596*** 0.316***(0.080) (0.073) (0.086) (0.062) (0.072) (0.029) (0.081) (0.021) (0.069) (0.036) (0.033) (0.023)

q25 0.017 0.047 0.221*** 0.179*** 0.131 0.109*** -0.006 0.207*** 0.158*** 0.498*** 0.634*** 0.384***(0.061) (0.031) (0.078) (0.024) (0.094) (0.021) (0.052) (0.029) (0.052) (0.032) (0.027) (0.028)

q50 0.114*** 0.128*** 0.397*** 0.060 0.217** 0.160*** -0.012 0.293*** 0.285*** 0.191*** 0.643*** 0.420***(0.042) (0.040) (0.109) (0.038) (0.086) (0.023) (0.043) (0.021) (0.048) (0.025) (0.022) (0.022)

q75 0.134** 0.243*** 0.638*** -0.096* 0.214** 0.164*** -0.038 0.388*** 0.356*** -0.016 0.643** 0.436***(0.054) (0.056) (0.177) (0.051) (0.091) (0.030) (0.086) (0.029) (0.081) (0.019) (0.042) (0.026)

q90 0.068 0.348*** 0.485** -0.280*** 0.001 0.140*** 0.062 0.578*** 0.591*** -0.184*** 0.670*** 0.497***(0.096) (0.061) (0.225) (0.053) (0.189) (0.033) (0.108) (0.058) (0.224) (0.030) (0.056) (0.023)

L1.CTF q10 0.793*** 0.594*** 0.553*** 1.304*** 0.802*** 0.866*** 0.154*** 0.118*** 0.121*** 0.355*** 0.872*** 0.783***(0.072) (0.043) (0.041) (0.121) (0.101) (0.030) (0.020) (0.024) (0.021) (0.067) (0.049) (0.020)

q25 0.721*** 0.541*** 0.485*** 1.276*** 0.899*** 0.900*** 0.399*** 0.201*** 0.094*** 0.988*** 1.219*** 0.903***(0.060) (0.037) (0.046) (0.040) (0.088) (0.020) (0.073) (0.015) (0.014) (0.046) (0.023) (0.018)

q50 0.673*** 0.576*** 0.534*** 1.337*** 1.136*** 1.072*** 0.500*** 0.252*** 0.121*** 1.324*** 1.311*** 1.083***(0.056) (0.043) (0.050) (0.056) (0.055) (0.025) (0.064) (0.016) (0.016) (0.039) (0.025) (0.014)

q75 0.865*** 0.689*** 0.834*** 1.566*** 1.545*** 1.334*** 0.612*** 0.277*** 0.214*** 1.508*** 1.421*** 1.244***(0.123) (0.062) (0.088) (0.060) (0.062) (0.038) (0.092) (0.022) (0.036) (0.036) (0.035) (0.021)

q90 1.336*** 0.910*** 1.299*** 1.771*** 1.536*** 1.549*** 0.865*** 0.380*** 0.429*** 1.577*** 1.526*** 1.401***(0.087) (0.068) (0.116) (0.074) (0.088) (0.050) (0.125) (0.025) (0.061) (0.059) (0.048) (0.023)

PD q10 0.024*** 0.014*** 0.006** 0.034*** 0.051*** 0.019*** 0.010 0.004*** 0.007*** 0.030*** 0.043*** 0.025***(0.005) (0.002) (0.003) (0.006) (0.006) (0.001) (0.006) (0.001) (0.002) (0.004) (0.004) (0.001)

q25 0.013*** 0.010*** 0.008*** 0.035*** 0.045*** 0.017*** 0.012*** 0.003*** 0.007*** 0.029*** 0.038*** 0.027***(0.004) (0.001) (0.002) (0.004) (0.004) (0.001) (0.003) (0.001) (0.002) (0.002) (0.003) (0.001)

q50 0.013*** 0.009*** 0.012*** 0.024*** 0.059*** 0.016*** 0.015** 0.004*** 0.004* 0.028*** 0.045*** 0.028***(0.004) (0.002) (0.003) (0.005) (0.009) (0.001) (0.005) (0.001) (0.002) (0.003) (0.004) (0.001)

q75 0.018*** 0.013*** 0.026*** 0.030*** 0.085*** 0.023*** 0.020*** 0.003 0.009** 0.033*** 0.062*** 0.034***(0.005) (0.002) (0.007) (0.005) (0.012) (0.001) (0.007) (0.002) (0.004) (0.004) (0.004) (0.002)

q90 0.012 0.021*** 0.048*** 0.048*** 0.149*** 0.035*** 0.012 0.000 0.025*** 0.062*** 0.088*** 0.044***(0.009) (0.004) (0.010) (0.008) (0.021) (0.002) (0.010) (0.004) (0.006) (0.005) (0.008) (0.003)

PC q10 -0.031 -0.016 0.002 -0.103** -0.065 -0.014** -0.042 -0.008 0.002 -0.075*** -0.082*** -0.063***(0.029) (0.015) (0.023) (0.045) (0.043) (0.006) (0.026) (0.008) (0.009) (0.019) (0.018) (0.010)

q25 -0.010 -0.008 -0.000 -0.058** -0.039 -0.013*** -0.037 -0.012** -0.005 -0.099*** -0.067*** -0.062***(0.021) (0.009) (0.018) (0.024) (0.040) (0.005) (0.026) (0.005) (0.009) (0.018) (0.015) (0.006)

q50 -0.015 -0.003 -0.021 -0.035 -0.077** -0.024*** -0.053** -0.022*** -0.001 -0.085*** -0.068*** -0.065***(0.017) (0.010) (0.015) (0.029) (0.037) (0.006) (0.024) (0.006) (0.009) (0.016) (0.017) (0.009)

q75 -0.044 -0.013 -0.048 -0.084** -0.071 -0.032*** -0.051 -0.029*** -0.031 -0.096*** -0.068*** -0.077***(0.030) (0.017) (0.033) (0.033) (0.044) (0.006) (0.038) (0.008) (0.019) (0.019) (0.014) (0.010)

q90 -0.059 -0.036 -0.091** -0.133*** -0.089* -0.048*** -0.013 -0.035* -0.086*** -0.139*** -0.115*** -0.090***(0.045) (0.025) (0.039) (0.034) (0.048) (0.011) (0.059) (0.020) (0.030) (0.032) (0.025) (0.010)

36

Table 9 - Continued: Quantile regression (QR): Firm-level returns to human capital and innovation, by country and industry


SIZE q10 0.023* 0.022*** 0.018* -0.069*** -0.039** 0.008* 0.003 -0.005 0.017*** -0.039*** -0.039*** -0.016***(0.014) (0.005) (0.010) (0.013) (0.015) (0.004) (0.012) (0.004) (0.006) (0.007) (0.006) (0.003)

q25 0.021*** 0.016*** 0.005 -0.081*** -0.038*** 0.000 0.005 -0.003 0.000 -0.052*** -0.030*** -0.024***(0.007) (0.003) (0.004) (0.007) (0.012) (0.003) (0.008) (0.003) (0.004) (0.005) (0.005) (0.002)

q50 0.022*** 0.005 0.004 -0.076*** -0.032*** -0.004 0.013** -0.009** -0.006 -0.047*** -0.035*** -0.034***(0.007) (0.004) (0.007) (0.008) (0.008) (0.003) (0.006) (0.004) (0.005) (0.004) (0.004) (0.003)

q75 0.036*** -0.002 -0.011 -0.065*** -0.041*** -0.010*** -0.005 -0.014*** -0.016*** -0.033*** -0.042*** -0.035***(0.012) (0.005) (0.012) (0.012) (0.011) (0.004) (0.021) (0.004) (0.005) (0.004) (0.004) (0.003)

q90 0.034* -0.015** -0.040** -0.055*** -0.055*** -0.018*** -0.006 -0.018*** -0.020 -0.030*** -0.040*** -0.043***(0.018) (0.007) (0.016) (0.017) (0.013) (0.005) (0.026) (0.006) (0.015) (0.009) (0.007) (0.004)

CAP q10 0.016 0.027*** 0.035*** 0.047*** 0.011 0.035*** 0.030** 0.085*** 0.070*** 0.169*** 0.141*** 0.139***(0.013) (0.006) (0.011) (0.012) (0.008) (0.005) (0.013) (0.004) (0.007) (0.007) (0.008) (0.005)

q25 0.026*** 0.035*** 0.034*** 0.079*** 0.020*** 0.045*** 0.194*** 0.055*** 0.081*** 0.106*** 0.075*** 0.124***(0.008) (0.004) (0.006) (0.006) (0.007) (0.003) (0.007) (0.012) (0.003) (0.006) (0.013) (0.005)

q50 0.045*** 0.044*** 0.043*** 0.088*** 0.020** 0.050*** 0.107*** 0.087*** 0.088*** 0.161*** 0.082*** 0.106***(0.008) (0.005) (0.008) (0.007) (0.008) (0.003) (0.013) (0.004) (0.005) (0.006) (0.005) (0.004)

q75 0.066*** 0.058*** 0.058*** 0.101*** -0.000 0.047*** 0.158*** 0.092*** 0.102*** 0.131*** 0.075*** 0.090***(0.008) (0.006) (0.011) (0.008) (0.007) (0.003) (0.0159) (0.005) (0.011) (0.007) (0.005) (0.004)

q90 0.056*** 0.074*** 0.073*** 0.085*** -0.013 0.045*** 0.163*** 0.106*** 0.108*** 0.129*** 0.068*** 0.077***(0.019) (0.008) (0.017) (0.009) (0.014) (0.005) (0.021) (0.008) (0.017) (0.008) (0.010) (0.005)

MAT q10 0.373*** 0.460*** 0.550*** 0.269*** 0.437*** 0.384*** 0.488*** 0.621*** 0.677*** 0.240*** 0.134*** 0.260***(0.035) (0.014) (0.019) (0.017) (0.031) (0.010) (0.035) (0.011) (0.008) (0.014) (0.006) (0.005)

q25 0.380*** 0.468*** 0.551*** 0.243*** 0.411*** 0.357*** 0.452*** 0.586*** 0.673*** 0.198*** 0.108*** 0.229***(0.022) (0.013) (0.015) (0.010) (0.025) (0.006) (0.027) (0.007) (0.006) (0.009) (0.003) (0.004)

q50 0.378*** 0.442*** 0.506*** 0.222*** 0.332*** 0.301*** 0.381*** 0.544*** 0.647*** 0.150*** 0.098*** 0.197***(0.021) (0.013) (0.020) (0.010) (0.021) (0.006) (0.018) (0.006) (0.008) (0.005) (0.003) (0.003)

q75 0.324*** 0.390*** 0.410*** 0.184*** 0.237*** 0.226*** 0.324*** 0.518*** 0.591*** 0.127*** 0.113*** 0.174***(0.039) (0.016) (0.023) (0.010) (0.012) (0.010) (0.029) (0.010) (0.010) (0.003) (0.004) (0.004)

q90 0.270*** 0.315*** 0.308*** 0.185*** 0.179*** 0.179*** 0.257*** 0.453*** 0.479*** 0.1029*** 0.145*** 0.160***(0.036) (0.015) (0.020) (0.010) (0.011) (0.008) (0.025) (0.010) (0.017) (0.005) (0.005) (0.003)

Notes: The dependent variable is ln(real labor productivity). Significance level of ***1%, **5%, *10%. Bootstrapped standard errors (20 replications). Quantileregressions additionally include GP, EAST (for DE ) and time dummies. Number of observations: See Table 7. Results are based on pooled simultaneous-quantileregressions for θ ∈ {0.05, 0.10, 0.20, 0.25, 0.30, 0.40, 0.50, 0.60, 0.70, 0.75, 0.80, 0.90, 0.95}. Results for other quantiles are available upon request.

37

Table 10: Impact of human capital, innovation, size, physical capital and material on industry productivity distribution, by country


HC Dispersion 0.769 0.677*** 0.485*** -3.289 0.242 0.201** 0.716 0.304*** 0.385*** -1.064*** 0.007 0.064**(0.680) (0.173) (0.162) (2.897) (0.388) (0.097) (2.075) (0.062) (0.154) (0.085) (0.041) (0.032)

Skewness -0.656 0.165 0.158 0.134 -1.070 -0.881 0.646 0.049 -0.291 -0.196** -0.945 -0.373(0.687) (0.366) (0.491) (0.255) (2.227) (0.657) (2.926) (0.233) (0.509) (0.084) (8.673) (0.612)

Kurtosis -0.977 1.179** 1.371*** -0.136 1.522 2.881* -1.511 4.037*** 3.655*** -0.775*** 145.3 15.55(1.113) (0.594) (0.498) (0.374) (2.786) (1.737) (6.553) (0.686) (1.458) (0.086) (879.6) (7.978)

L1.CTF Dispersion 0.091* 0.121*** 0.265*** 0.102*** 0.264*** 0.194*** 0.211** 0.159*** 0.390*** 0.208*** 0.077*** 0.159***(0.051) (0.032) (0.051) (0.026) (0.044) (0.012) (0.095) (0.038) (0.075) (0.019) (0.013) (0.010)

Skewness 1.667** 0.522* 0.720*** 0.584 0.265* 0.207*** 0.049 -0.349 0.554*** -0.294*** 0.093 -0.059(0.685) (0.269) (0.164) (0.362) (0.143) (0.079) (0.411) (0.304) (0.176) (0.092) (0.189) (0.070)

Kurtosis 14.797 10.138*** 5.297*** 10.606*** 3.618*** 5.571*** 5.729** 7.021*** 4.547*** 4.710*** 13.17*** 6.402***(9.483) (2.802v (1.064) (2.996) (0.624) (0.399) (2.783) (1.759) (1.052) (0.451) (2.270) (0.405)

PD Dispersion 0.166 0.134** 0.526*** -0.079 0.309*** 0.159*** 0.252 0.089 0.161 0.059 0.238*** 0.109***(0.141) (0.054) (0.147) (0.082) (0.072) (0.039) (0.196) (0.313) (0.206) (0.052) (0.044) (0.026)

Skewness 0.918 1.448 0.546* -3.146 0.275 1.183*** 0.029 -3.131 3.358 1.779 0.436** 0.824***(1.281) (0.950) (0.288) (3.504) (0.398) (0.376) (1.166) (14.56) (4.609) (0.301) (0.225) (0.320)

Kurtosis 7.045 11.204** 2.988** -16.121 5.024*** 8.545*** 2.770 8.393 12.49 25.11 5.498*** 10.23***(6.817) (4.544) (1.182) (16.801) (1.585) (2.138) (2.616) (27.41) (17.09) (23.05) (1.011) (2.407)

PC Dispersion 0.642 0.234 0.988 0.183 0.290 0.422*** 0.161 0.434** 0.733** -0.017 0.006 0.110**(0.638) (0.507) (0.719) (0.200) (0.422) (0.133) (0.467) (0.182) (0.407) (0.885) (0.084) (0.047)

Skewness 0.677 3.053 0.151 2.737 -1.392 -0.165 -1.305 -0.190 1.320 -7.179 -2.289 0.604(0.834) (7.802) (0.698) (3.765) (2.531) (0.393) (5.615) (0.536) (0.827) (47.34) (52.60) (0.684)

Kurtosis 2.613 10.251 1.851 9.078 4.805 3.212*** 3.864 2.426* 3.195** -64.11 258.9 10.01**(2.618) (25.493) (1.360) (10.972) (6.875) (1.228) (10.28) (1.393) (1.645) (444.5) (3850) (4.551)

SIZE Dispersion 0.254 -1.215 2.518 -0.110 0.041 1.051 42.02 0.615*** 1.021* -0.221*** 0.167** 0.177***(0.157) (0.789) (3.918) (0.102) (0.158) (0.675) (4642) (0.230) (0.570) (0.069) (0.087) (0.054)

Skewness 0.969 -0.185 0.844 0.342 4.660 0.180 2.887 -0.133 0.293 0.458 0.115 -0.881(0.827) (0.311) (0.772) (0.935) (18.145) (0.399) (4.463) (0.484) (0.488) (0.350) (0.389) (0.624)

Kurtosis 3.948 -0.398 1.411 -7.682 29.020 0.961 0.388 2.138*** 0.188 -3.651*** 6.568** 5.659***(2.796) (0.464) (1.371) (7.228) (113.736) (0.714) (3.318) (0.860) (0.885) (1.335) (3.452) (1.635)

CAP Dispersion 0.431*** 0.249*** 0.263** 0.125*** -1.033 0.020 0.486*** 0.062*** 0.127*** -0.194*** -0.170*** -0.162***(0.117) (0.045) (0.111) (0.041) (0.690) (0.039) (0.088) (0.023) (0.048) (0.026) (0.034) (0.018)

Skewness 0.023 0.178 0.226 0.146 0.957 -3.698 -0.014 0.017 0.196 -0.051 -0.549* -0.047(0.261) (0.247) (0.392) (0.495) (0.927) (7.459) (0.176) (0.434) (0.328) (0.114) (0.295) (0.149)

Kurtosis 1.807** 4.347*** 4.468** 5.838*** 0.102 44.227 1.872*** 17.633*** 7.753*** -4.737*** -6.749*** -6.217***(0.794) (0.933) (1.992) (1.965) (0.918) (88.969) (0.320) (6.460) (3.032) (0.709) (1.365) (0.691)

MAT Dispersion -0.080* -0.091*** -0.147*** -0.138*** -0.268*** -0.225*** -0.165*** -0.061*** -0.065*** -0.218*** 0.023 -0.136***(0.047) (0.013) (0.021) (0.029) (0.023) (0.019) (0.051) (0.008) (0.007) (0.021) (0.021) (0.010)

Skewness 0.912* 0.335* 0.362*** 0.294 0.091 0.134** -0.107 -0.225** 0.371*** -0.363*** 5.097 -0.282***(0.514) (0.181) (0.131) (0.274) (0.157) (0.054) (0.180) (0.105) (0.101) (0.095) (4.841) (0.072)

Kurtosis -11.466* -9.976*** -6.061*** -7.725*** -3.549*** -4.293*** -5.839*** -15.849*** -14.021*** -4.827*** 55.29 -7.679***(6.414) (1.455) (0.845) (1.559) (0.298) (0.317) (1.806) (1.952) (1.506) (0.491) (150.81) (0.593)

Notes: Significance level of ***1%, **5%, *10%. Tests are based on estimation results in Table 9.

38

Table 11: FE quantile regression (FEQR): Firm-level returns to human capital and innovation, by country and industry


HC q10 -0.194*** -0.118*** 0.141 0.009 0.161 -0.038 0.240*** 0.057* -0.303** 0.018 -0.048 0.001(0.052) (0.038) (0.117) (0.026) (0.120) (0.023) (0.061) (0.031) (0.075) (0.023) (0.034) (0.024)

q25 -0.129*** -0.096** 0.178*** 0.010 0.212*** 0.003 0.201*** 0.144*** -0.190*** -0.013* 0.013 0.028**(0.037) (0.039) (0.062) (0.018) (0.053) (0.015) (0.046) (0.020) (0.036) (0.014) (0.020) (0.014)

q50 -0.083*** -0.033** 0.156*** 0.032*** 0.188*** 0.038*** 0.198*** 0.158*** -0.176*** -0.023** 0.041*** 0.056***(0.017) (0.017) (0.023) (0.011) (0.026) (0.012) (0.032) (0.02) (0.033) (0.010) (0.014) (0.011)

q75 -0.022 -0.001 0.142** 0.058*** 0.160*** 0.093*** 0.167*** 0.217*** -0.151*** -0.035** 0.065*** 0.083***(0.039) (0.037) (0.057) (0.015) (0.060) (0.014) (0.044) (0.017) (0.038) (0.015) (0.017) (0.012)

q90 0.058 0.135** 0.200 0.064*** 0.142 0.136*** 0.144 0.298*** -0.118** -0.060** 0.094*** 0.119***(0.067) (0.068) (0.122) (0.025) (0.144) (0.026) (0.103) (0.029) (0.057) (0.026) (0.031) (0.022)

L1.CTF q10 0.496*** 0.491*** 0.266*** 0.789*** 0.586*** 0.525*** 0.227*** 0.183*** 0.185*** 0.590*** 0.513*** 0.403***(0.064) (0.031) (0.035) (0.039) (0.046) (0.023) (0.034) (0.011) (0.015) (0.022) (0.017) (0.010)

q25 0.513*** 0.482*** 0.252*** 0.819*** 0.606*** 0.532*** 0.218*** 0.188*** 0.188*** 0.599*** 0.507*** 0.410***(0.033) (0.018) (0.021) (0.028) (0.039) (0.013) (0.036) (0.009) (0.012) (0.022) (0.020) (0.008)

q50 0.523*** 0.474*** 0.257*** 0.869*** 0.657*** 0.539*** 0.182*** 0.204*** 0.202*** 0.625*** 0.539*** 0.436***(0.025) (0.014) (0.014) (0.013) (0.015) (0.009) (0.033) (0.008) (0.012) (0.017) (0.010) (0.006)

q75 0.496*** 0.462*** 0.239*** 0.853*** 0.652*** 0.539*** 0.150*** 0.219*** 0.193*** 0.662*** 0.580*** 0.479***(0.038) (0.020) (0.025) (0.024) (0.027) (0.011) (0.044) (0.011) (0.014) (0.031) (0.015) (0.008)

q90 0.554*** 0.477*** 0.226*** 0.883*** 0.718*** 0.562*** 0.164** 0.237*** 0.202*** 0.666*** 0.621*** 0.496***(0.050) (0.026) (0.035) (0.043) (0.074) (0.020) (0.068) (0.016) (0.022) (0.036) (0.024) (0.011)

PD q10 0.005 0.015*** 0.008** 0.030*** 0.028*** 0.019*** 0.006 0.008*** 0.019*** 0.023*** 0.031*** 0.021***(0.004) (0.002) (0.003) (0.003) (0.004) (0.001) (0.005) (0.002) (0.003) (0.002) (0.003) (0.001)

q25 0.006** 0.016*** 0.010*** 0.027*** 0.023*** 0.019*** 0.011*** 0.006*** 0.014*** 0.020*** 0.025*** 0.019***(0.002) (0.001) (0.002) (0.002) (0.003) (0.001) (0.004) (0.001) (0.001) (0.001) (0.002) (0.001)

q50 0.007*** 0.016*** 0.010*** 0.027*** 0.023*** 0.018*** 0.013*** 0.007*** 0.014*** 0.021*** 0.025*** 0.019***(0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001)

q75 0.004 0.017*** 0.011*** 0.027*** 0.016*** 0.018*** 0.018*** 0.006*** 0.014*** 0.022*** 0.025*** 0.020***(0.003) (0.001) (0.002) (0.002) (0.002) (0.001) (0.004) (0.001) (0.001) (0.002) (0.002) (0.001)

q90 -0.000 0.020*** 0.011*** 0.028*** 0.019** 0.020*** 0.021*** 0.005*** 0.016*** 0.028*** 0.026*** 0.022***(0.005) (0.002) (0.003) (0.003) (0.008) (0.002) (0.006) (0.001) (0.003) (0.004) (0.003) (0.001)

PC q10 -0.015 -0.023** 0.011 -0.012 -0.087*** -0.015** -0.024 0.009 -0.003 -0.005 -0.066** -0.029***(0.024) (0.011) (0.013) (0.018) (0.028) (0.008) (0.028) (0.008) (0.012) (0.014) (0.017) (0.007)

q25 -0.016 -0.031*** 0.001 -0.004 -0.046*** -0.020*** -0.025 0.001 -0.001 -0.011 -0.042*** -0.023***(0.012) (0.006) (0.009) (0.010) (0.015) (0.004) (0.022) (0.005) (0.006) (0.009) (0.010) (0.004)

q50 -0.009 -0.031*** -0.003 -0.008 -0.034*** -0.019*** -0.018 0.000 -0.002 -0.021*** -0.036*** -0.023***(0.008) (0.005) (0.007) (0.007) (0.010) (0.003) (0.014) (0.003) (0.006) (0.008) (0.005) (0.003)

q75 -0.004 -0.034*** -0.009 -0.012 -0.008 -0.017*** -0.038* -0.001 -0.007 -0.023** -0.043*** -0.029***(0.015) (0.006) (0.010) (0.009) (0.013) (0.005) (0.020) (0.005) (0.007) (0.010) (0.007) (0.004)

q90 -0.008 -0.032*** -0.017 -0.014 0.004 -0.024*** -0.004 -0.004 -0.017 -0.031* -0.043*** -0.031***(0.019) (0.012) (0.015) (0.016) (0.025) (0.009) (0.033) (0.007) (0.011) (0.016) (0.013) (0.007)

Notes: Sample: �rms with 2 or more observations (DE: 8,117; NL: 15,427 observations). The dependent variable is ln(real labor productivity). Included but not reported are SIZE,CAP, MAT, GP, EAST (for DE), time dummies and industry dummies (only for the total sample).

39

Table 12: Impact of human capital and innovation on industry productivity distribution, by country


HC Dispersion -0.712* -0.986 -0.114 0.705 -0.141 0.930*** -0.094 0.203*** -0.113 0.447 0.668* 0.497***(0.417) (0.754) (0.220) (0.437) (0.192) (0.287) (0.135) (0.031) (0.122) (0.432) (0.395) (0.173)

Skewness 0.137 -0.335 -0.221 0.089 0.078 0.226 0.815 0.616* 0.305 0.085 -0.074 -0.018(0.460) (0.483) (2.150) (0.452) (1.397) (0.209) (1.997) (0.356) (1.341) (0.766) (0.470) (0.247)

Kurtosis -1.259 0.170 -9.376 1.532 -5.783 1.097** -11.13 4.844*** -10.92 1.962 0.868 2.168***(1.024) (0.843) (18.453) (1.074) (7.997) (0.458) (16.91) (1.567) (11.86) (2.321) (1.060) (0.893)

L1.CTF Dispersion -0.017 -0.022 -0.028 0.020 0.037 0.007 -0.185 0.076*** 0.011 0.050** 0.067*** 0.079***(0.039) (0.024) (0.058) (0.018) (0.031) (0.012) (0.126) (0.025) (0.045) (0.024) (0.018) (0.011)

Skewness 2.207 0.158 1.719 -1.972 -1.202 -0.887 -0.039 -0.026 -5.223 0.166 0.116 0.234(5.237) (1.229) (3.807) (1.966) (1.075) (2.160) (0.726) (0.412) (21.65) (0.525) (0.284) (0.153)

Kurtosis -60.875 -47.057 -36.311 49.843 28.265 150.865 -6.097 12.90*** 94.66 19.32** 15.08*** 12.37***(138.899) (51.303) (76.384) (45.389) (23.153) (260.788) (4.407) (4.220) (369.5) (9.248) (3.967) (1.724)

PD Dispersion -0.216 0.016 0.049 0.007 -0.159* -0.019 0.266* -0.039 0.008 0.048 -0.008 0.032(0.341) (0.040) (0.103) (0.043) (0.083) (0.026) (0.157) (0.083) (0.055) (0.052) (0.036) (0.025)

Skewness 1.732 0.076 0.603 2.275 1.197 -1.129 0.451 2.925 -1.878 0.098 1.260 0.430(2.816) (2.909) (2.393) (16.904) (0.882) (2.366) (0.715) (6.666) (15.53) (1.250) (7.386) (0.836)

Kurtosis -2.139 65.559 17.797 151.569 -7.438* -54.629 3.513* -25.41 153.3 24.97 -138.9 33.73(5.091) (163.218) (37.428) (924.778) (4.361) (71.780) (2.181) (54.83) (1056) (27.19) (612.1) (26.58)

PC Dispersion -0.571 0.044 1.326 0.506 -0.712* -0.090 0.199 33.88 0.857 0.373 0.019 0.132(1.112) (0.117) (2.853) (0.900) (0.394) (0.142) (0.359) (3990) (1.329) (0.327) (0.121) (0.095)

Skewness -0.247 0.748 0.129 0.093 0.414 0.434 2.158 0.482 0.564 -0.659 7.371 0.861(1.544) (3.272) (1.217) (1.352) (0.552) (1.798) (4.298) (3.091) (1.662) (1.081) (48.60) (0.990)

Kurtosis -2.077 19.577 0.499 3.214 -2.197* -12.097 2.237 -2.013 3.104 2.850 66.58 8.759(3.970) (51.442) (2.147) (5.439) (1.296) (18.721) (5.117) (7.332) (4.217) (2.601) (414.7) (6.388)

Notes: Signi�cance level of ***1%, **5%, *10%. Tests are based on FE quantile regression estimation results in Table 11.

40

Graph 1: Distribution of human capital intensity, by country and industry

01

23

45

67

89

0 .2 .4 .6 .8 1share of highskil led emp.

HT MT LT KIS OS

GERMANY

41

Graph 2: Innovation performance distribution, by country and industry

0.1

.2.3

.4.5

2 0 2 4 6 8ln(real innovative sales per emp.)

HT MT LT KIS OS

GERMANY

42

Graph 3: Productivity distribution, by country and industry

0.2

.4.6

.81

2 0 2 4 6 8 10ln(real labor productivity)

HT MT LT KIS OS

GERMANY

43

Graph 4: Average and quantile impact of human capital on productivity, by country and industry

.10

.1.2

.3pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

TOTAL

.50

.5pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

HT

.20

.2.4

.6.8

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

MT

.50

.51

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

LT

.50

.5pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

KIS

.50

.51

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

OS

GERMANY

Note: Solid lines present coeffi cient estimates and 95% confidence intervals of quantile regressions. For comparison, dashed lines mark coeffi cient estimates and 95% confidenceintervals of the OLS regression.

44

Graph 5: Average and quantile impact of product innovation on productivity, by country and industry

.01

.02

.03

.04

.05

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

TOTAL

.01

0.0

1.0

2.0

3.0

4pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

HT

0.0

1.0

2.0

3.0

4pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

MT

0.0

5.1

.15

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

LT

.01

.02

.03

.04

.05

.06

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

KIS

.50

.51

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

OS

GERMANY


45

Graph 6: Average and quantile impact of process innovation on productivity, by country and industry

.15

.1.0

50

prod

uctiv

ity e

ffect

0 20 40 60 80 100quantile

TOTAL

.15

.1.0

50

.05

prod

uctiv

ity e

ffect

0 20 40 60 80 100quantile

HT

.15

.1.0

50

.05

prod

uctiv

ity e

ffect

0 20 40 60 80 100quantile

MT

.2.1

5.1

.05

0.0

5pr

oduc

tivity

effe

ct

0 20 40 60 80 100quantile

LT

.25

.2.1

5.1

.05

0pr

oduc

tivity

effe

ct

0 20 40 60 80 100quantile

KIS

.3.2

.10

.1pr

oduc

tivity

effe

ct

0 20 40 60 80 100quantile

OS

GERMANY


46

Graph 7: FE: Average and quantile impact of human capital on productivity, by country and industry

.10

.1.2

.3pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

TOTAL

.3.2

.10

.1.2

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

HT

.2.1

0.1

.2.3

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

MT

.50

.51

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

LT

.05

0.0

5.1

.15

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

KIS

.20

.2.4

.6.8

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

OS

GERMANY

Note: Solid lines present coeffi cient estimates and 95% confidence intervals of quantile regressions. For comparison, dashed lines mark coeffi cient estimates and 95% confidenceintervals of the FE regression.

47

Graph 8: FE: Average and quantile impact of product innovation on productivity, by country and industry

.015

.02

.025

.03

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

TOTAL

.01

0.0

1.0

2.0

3pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

HT

.01

.015

.02

.025

.03

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

MT

.005

0.0

05.0

1.0

15pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

LT

.02

.025

.03

.035

.04

.045

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

KIS

0.0

1.0

2.0

3.0

4.0

5pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

OS

GERMANY


48

Graph 9: FE: Average and quantile impact of process innovation on productivity, by country and industry

.06

.04

.02

0pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

TOTAL

.15

.1.0

50

.05

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

HT

.08

.06

.04

.02

0.0

2pr

oduc

tivity

effe

ct

0 20 40 60 80 100quanti le

MT

.1.0

50

.05

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

LT

.1.0

50

.05

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

KIS

.15

.1.0

50

.05

prod

uctiv

ity e

ffect

0 20 40 60 80 100quanti le

OS

GERMANY


49

Appendix A : Measurement details

A.1 Measurement of the human capital variable in the Dutch data

In order to de�ne the education type of employees in the matched (CIS\PS)-enterprises, we built a matchedemployer-employee microdata set by merging our enterprise data with the Social Statistics Database (SSB). Thepopulation of interest consists of individuals aged 15-65 covering the period 1999-2008.1 Table A.1 reports thenumber of employees (N), the number of enterprises, and the median number of employees per enterprise for eachyear in manufacturing and services in the matched employer-employee data.

Table A.1: Panel structure of matched employer-employee microdata set - 1999-2008

MANUFACTURING SERVICES

Year # emp.% emp.

in Educ.# �rms

# emp.�rm

a)# employ.

% emp.

in Educ.# �rms

# emp.�rm

a)

1999 768; 844 19:2 9; 452 30 1; 749; 492 30:5 14; 320 29

2000 759; 266 20:1 9; 284 31 1; 796; 189 32:0 14; 192 31

2001 745; 032 20:4 9; 244 31 1; 760; 933 32:4 14; 382 32

2002 705; 867 21:1 9; 048 30 1; 729; 602 33:3 14; 417 32

2003 677; 188 22:6 8; 842 30 1; 669; 277 34:8 14; 236 31

2004 648; 995 24:3 8; 675 29 1; 667; 713 36:6 14; 086 31

2005 626; 966 26:2 8; 429 29 1; 664; 649 39:3 13; 766 31

2006 623; 756 29:2 8; 074 30 1; 686; 114 42:8 13; 253 32

2007 614; 249 31:4 7; 875 31 1; 720; 888 45:4 12; 987 32

2008 611; 725 33:8 7; 496 31 1; 722; 096 47:5 12; 194 33

Note: a) Median value.

The education type of each employee is determined in two stages. In the �rst stage, the matched employer-employee microdata set is linked to the Education database which provides the highest level of education at-tained by an individual. The education type is based on a 2-digit SOI-code (Dutch education classi�cation:Standaard Onderwijsindeling) and is converted to the ISCED classi�cation (International Standard Classi�cationof Education). Table A.2 provides details on the Dutch education system and on the mapping between the SOIand the ISCED classi�cations.

1We select the period 1999-2008 since this period is covered in the Education database (see supra).

50

Table A.2: The Dutch education system

Dutch education systemSOIcode

ISCEDcode

3-skilltype

4-skilltype

Pre-primary education, age 4-5 0

Primary education, age 6-12 20 1 LS LS

Lower secondary education, age 13-16:

- vocational: MBO (level 1), VMBO (grade 3-4)

- general: VMBO (grade 1-2), HAVO/VWO (grade 1-3),

MAVO (grade 1-4)

31-33 2 LS LS

Higher secondary education, age 17-18:

- vocational: MBO (level 2-4)

- general: HAVO/VWO (grade 4-6)

41-42 3 MS LMS

Post-secondary, non-tertiary education, age > 19:

- MBO (level 4)

- 1-year HBO

43 4 MS HMS

Tertiary education, type B: 2-3 year HBO 51-52 5B HS HMS

Tertiary education, type A:

- 4-6 year HBO- WO and HBO Bachelor, WO Master

53

60

5A

5A

HS

HS

HS

HS

Advanced research quali�cation: AIO, OIO, WO-Ph.D. 6

On the basis of the ISCED-codes, we characterize two decompositions of the workforce which are reported in the lasttwo columns of Table A.2. Following Antenbrink et al. (2005), the �rst decomposition splits the workforce into threeskill types (low-skilled (LS), medium-skilled (MS) and high-skilled (HS)). In line with O�Mahony et al. (2008),the second decomposition further re�nes the middle type into low-medium-skilled (LMS) and high-medium-skilled(HMS) types. The third and seventh columns in Table A.1 report the fraction of employees that are observed inthe Education database in manufacturing and services respectively. The fraction lies in the [19:2%-33:8%]-range formanufacturing and in the [30:5%-47:5%]-range for services.

In the second stage, we determine the skill type of employees who are not observed in the Education database.For that purpose, we estimate a reverse Mincer equation. More speci�cally, we estimate an ordered probit modelto predict each individual�s skill type (LS; MS; HS) based on individual and �rm characteristics in the matchedemployer-employee microdata for each year during the period 1999-2008. The ordered probit model is built arounda latent regression equation:

Skill�j(i) = xj�+ zi� + �j (A.1)

where Skill�j(i) is the skill type of individual i working in enterprise j, xj a vector of the individual�s familybackground and labor market characteristics, zi a vector of enterprise characteristics and �j a normally distributederror term. We do not observe the latent variable Skill�j(i). However, the observed skill type can be modeled in thefollowing way:

Skillj(i) = l if cl�1 � Skill�j(i) < cl (A.2)

where l = 1; 2; 3 are the three skill types and cl are the cut-o¤ levels in the ordered probit model. To predict skilloutcomes, we use the following explanatory variables: age, age squared, tenure, tenure squared, ln(yearly gross

51

wage), ln(yearly working hours), 11 province dummies capturing the location of the individual2 , sex dummy (0 =female, 1 = male), marital status dummy (0 = married/widowed/divorced/registered partnership, 1 = married),birth country dummy (0 = other than the Netherlands (NL), 1 = NL), birth country father dummy (0 = otherthan NL, 1 = NL), birth country mother dummy (0 = other than NL, 1 = NL), 6 size class dummies3 and 20industry dummies4 . The estimation sample is restricted to individuals aged 15-65 with wage and working timevalues within the [p1-p99]-range.

Table A.3 presents the yearly skill composition of the workforce in manufacturing and services. The �rst percentagein each column refers to the proportion of respectively low-skilled, medium-skilled and high-skilled employees basedon the Education Database, i.e. the education (and hence skill) type for these individuals is observed. The secondpercentage in each column �put in square brackets�corresponds to the skill composition based on predicted skilloutcomes.5 The match between the observed and the predicted skill type for individuals in the Education Databaselies in the [58%-65%]-range in both manufacturing and services.6 Focusing on the skill composition in squarebrackets, we observe a slight decrease in the proportion of low-skilled employees and a considerable decrease inthe proportion of medium-skilled employees over time in both manufacturing and services which translates into asigni�cant increase in the proportion of high-skilled employees over time. The latter appears to be more pronouncedin manufacturing.

Table A.3: Skill composition of the workforce in matched employer-employee microdata set - 1999-2008

MANUFACTURING SERVICESYear % LS % MS % HS % LS % MS % HS

1999 25:0 [21:7] 43:1 [59:9] 31:8 [18:4] 22:0 [16:3] 46:1 [55:8] 32:0 [28:1]

2000 25:4 [21:7] 41:7 [58:5] 32:9 [19:8] 23:4 [16:9] 44:5 [53:8] 32:1 [29:3]

2001 24:3 [21:5] 41:3 [58:0] 34:3 [20:5] 22:6 [16:3] 44:3 [53:2] 33:1 [30:5]

2002 24:2 [21:7] 40:0 [56:1] 35:8 [22:2] 22:8 [16:7] 42:9 [51:3] 34:2 [32:0]

2003 25:7 [23:1] 37:9 [52:2] 36:4 [24:7] 25:4 [17:7] 40:5 [48:6] 34:1 [33:6]

2004 26:0 [25:2] 37:4 [49:1] 36:6 [25:7] 26:4 [18:2] 40:0 [47:5] 33:5 [34:3]

2005 25:9 [24:3] 37:7 [49:4] 36:5 [26:3] 26:2 [17:5] 40:6 [47:2] 33:2 [35:2]

2006 24:8 [23:0] 37:4 [48:8] 37:8 [28:2] 27:0 [18:9] 40:6 [46:7] 32:4 [34:4]

2007 26:0 [24:0] 37:8 [49:1] 36:2 [26:9] 27:9 [19:8] 40:7 [47:0] 31:3 [33:1]

2008 25:9 [24:1] 38:2 [49:4] 35:8 [26:4] 27:9 [20:1] 41:2 [47:8] 31:0 [32:1]

TOTALa) 25:5 [23:0] 38:0 [50:8] 36:0 [25:2] 25:8 [17:6] 40:9 [48:2] 32:7 [32:6]

Note: a) Median value.

2The 12 provinces are Groningen (reference), Friesland, Drenthe, Overijssel, Flevoland, Gelderland, Utrecht, Noord-Holland, Zuid-Holland, Zeeland, Noord-Brabant and Limburg.

3The 7 size classes are de�ned as follows: size class = 1 if the number of employees (L) < 10 (reference); size class = 2 if L 2 [10; 20[,size class = 3 if L 2 [20; 50[, size class = 4 if L 2 [50; 100[, size class = 5 if L 2 [100; 200[, size class = 6 if L 2 [200; 500[ and size class= 7 if L � 500.

4The 11 manufacturing industries are food, textiles, wood, chemicals, plastics, glass, metal, machinery, electrical engineering, vehicles,furniture/recycling and the 10 services industries are wholesale, transport, telecommunication, computer, technical services, consultancy,other business related services, renting, retail and R&D services.

5Evidently, we take the observed skill type for individuals in the Education Database. The predicted skill type is used for theremaining individuals.

6Details on the ordered probit estimates are not reported but available upon request.

52

We applied the same procedure to determine the skill type for each employee in the matched employer-employeemicrodata set based on the 4-skill type decomposition (see supra).7

As noted above, we performed the ordered probit regressions on a yearly basis. To investigate the stability ofan individual�s (observed or predicted) skill type over the considered period (1999-2008), we compared the skilltype of an individual in the �rst year of observation to her skill type in the last year of observation. Focusingon manufacturing, our unbalanced panel consists of 1; 470; 982 individuals over the period 1999-2008. The skilltype is observed for 31:1% of the individuals. Considering the subsample of individuals for which the skill typeis observed, 34:8% of the individuals belong to the low-skilled type, 38:1% to the medium-skilled type and 27:1%to the high-skilled type. Considering the total sample of individuals (for which the skill type is either observed orpredicted), the corresponding shares are 24:3%, 51:9% and 23:9%. The number of observations per individual is 2for the �rst quartile of individuals, 3 for the second quartile and 8 for the third quartile.8 Restricting the sample toindividuals having at least two observations, we observe that the skill type is unchanged for 69:1% of the individualswhereas 14:6% of the individuals experience skill upgrading and 16:4% skill downgrading. Focusing on services,our unbalanced panel consists of 4; 865; 343 individuals over the period 1999-2008. The skill type is observed for42:2% of the individuals. Considering the subsample of individuals for which the skill type is observed, 41:4%of the individuals are low-skilled, 38:7% medium-skilled and 19:9% high-skilled. Considering the total sample ofindividuals, the corresponding shares are 26:1%, 49:0% and 24:9%. The number of observations per individual is 1for the �rst quartile of individuals, 3 for the second quartile and 5 for the third quartile.9 Restricting the sample toindividuals having at least two observations, we observe that the skill type is unchanged for 66:6% of the individualswhereas 23:2% of the individuals experience skill upgrading and 10:2% skill downgrading. Since no clear patterncan be discerned in the skill type of the skill-downgrading category in both manufacturing and services, we decidedto leave the skill type of these individuals unchanged.

Finally, we determine the share of each skill type for each matched (CIS\PS)-enterprise by aggregating to theenterprise level.10 Table A.4 reports the means, standard deviations and quartile values of the skill types �de�nedas shares lying in the [0; 1]-range�in manufacturing and services. We further break down manufacturing and servicesinto �ve industries according to the OECD (2001) classi�cation: High-technology manufacturing (HT ), Medium-technology manufacturing (MT ), Low-technology manufacturing (LT ), Knowledge-intensive services (KIS) andOther services (OS).

7Details are not provided but available upon request.8Putting the number of individuals between brackets and the number of observations between square brackets, the structure of the

manufacturing data is given by: (333,076) [1], (242,420) [2], (163,997) [3], (103,604) [4], (83,037) [5], (71,751) [6], (75,460) [7], (71,246)[8], (86,136) [9], (240,255) [10]. The total number of observations is 6,845,976.

9Putting the number of individuals between brackets and the number of observations between square brackets, the structure of theservices data is given by: (1,300,050) [1], (1,015,217) [2], (677,490) [3], (476,719) [4], (335,782) [5], (247,174) [6], (205,536) [7], (174,679)[8], (172,800) [9], (259,896) [10]. The total number of observations is 17,422,128.10 Information on the skill decomposition of the workforce is missing for about 5% of the matched (CIS\PS)-enterprises.

53

Table A.4: Skill composition of the workforce in enterprise data set - 1999-2008

Variables Mean Sd. Q1 Q2 Q3 N

MANUFACTURING

LS 0.266 0.141 0.160 0.250 0.354 22 614

MS 0.557 0.128 0.476 0.556 0.641 22 883

HS 0.180 0.153 0.069 0.140 0.254 23 225

HTLS 0.139 0.089 0.071 0.118 0.190 1 549

MS 0.480 0.158 0.387 0.489 0.583 1 619

HS 0.387 0.205 0.237 0.362 0.522 1 644

MTLS 0.258 0.135 0.156 0.241 0.344 14 557

MS 0.560 0.122 0.480 0.557 0.639 14 738

HS 0.184 0.145 0.077 0.149 0.264 14 961

LTLS 0.313 0.142 0.210 0.293 0.402 6 508

MS 0.570 0.127 0.486 0.569 0.658 6 526

HS 0.119 0.099 0.044 0.100 0.170 6 620

SERVICES

LS 0.170 0.122 0.074 0.149 0.240 30 787

MS 0.518 0.186 0.385 0.545 0.656 33 766

HS 0.317 0.256 0.101 0.250 0.510 35 417

KISLS 0.141 0.130 0.041 0.094 0.214 12 319

MS 0.418 0.193 0.258 0.397 0.571 14 713

HS 0.439 0.290 0.152 0.493 0.692 15 901

OSLS 0.189 0.113 0.107 0.173 0.250 18 468

MS 0.596 0.137 0.506 0.602 0.692 19 053

HS 0.217 0.167 0.082 0.186 0.320 19 516

From Table A.4, it follows that the median proportion of high-skilled employees (HS) is about 14% in manufacturing.We observe considerable heterogeneity across industries: the median HS ranges from 10% in Low-technologymanufacturing industries to 36:2% in High-technology manufacturing industries. The median HS amounts to 25%in services, ranging from 18:6% in Other services to 49:3% in Knowledge-intensive services.

A.2 Measurement of closeness to the technological frontier variable in the Dutch data

A.2.1 Closeness-to-frontier variable based on real labor productivity

In order to de�ne our main closeness-to-the-technological-frontier variable which is based on real labor productiv-ity (CTFit�1), we consider the largest possible population of enterprises from the Production Surveys. After somecleaning and trimming on nominal labor productivity levels and growth rates to eliminate outliers and anomalies, we

54

have an unbalanced panel of 381; 546 observations corresponding to 130; 893 enterprises (35% in manufacturing and65% in services) over the period 1998-2008. 1:7% of the enterprises belong to High-technology manufacturing, 12:3%to Medium-technology manufacturing, 13:8% to Low-technology manufacturing, 31:2% to Knowledge-intensive ser-vices and 41:1% to Other services.

Table A.5: Panel structure of PS sample - 1998-2008

# consecutive years # �rms

� 2 74; 378

� 3 32; 114

� 4 22; 714

� 5 17; 310

� 6 12; 990

A.2.2 Closeness-to-frontier variable based on total factor productivity

In the robustness check using total factor productivity (TFP ) as the dependent variable, we include as a covariatethe one-year lagged value of the closeness-to-the-technological-frontier variable which is based on estimates of total

factor productivity�CTF TFP

it�1�. We measure the latter as CTFTFPit as 1�DTFTFPit = 1�

�[TFPFt�[TFP it

[TFPFt

�=

[TFP it

[TFPFt

where [TFP of the technological frontier �rm F is proxied by the 95% percentile value of [TFP at the NACE 3-digit industry level. The data that are used to estimate TFP of the technological frontier F stem from the largestpossible population of enterprises from the Production Surveys. After some cleaning and trimming on nominal laborproductivity levels and growth rates to eliminate outliers and anomalies and restricting the population to enterpriseshaving at least two consecutive years, our estimation sample consists of 292;770 observations corresponding to 74;378enterprises (40:5% in manufacturing and 59:5% in services) spanning the period 1998-2008. 2:1% of the enterprisesbelong to High-technology manufacturing, 16:8% to Medium-technology manufacturing, 19:6% to Low-technologymanufacturing, 22% to Knowledge-intensive services and 39:4% to Other services.

55

A.3 Breakdown of manufacturing and services according to technological intensity

Table A.6: Breakdown of manufacturing and services according to technological intensityNACE Rev. 1.1 codes

MANUFACTURING

High-technology manufacturing (HT)

24.4 Pharmaceuticals, medicinal chemicals and botanical products30 Offi ce machinery and computers32 Radio, television and communication equipment and apparatus33 Medical, precision and optical instruments, watches and clocks35.3 Aircraft and spacecraft

Medium-technology manufacturing (MT)

23 Coke, refined petroleum products and nuclear fuel

24 Chemicals and chemical products, excluding 24.425 to 28 Rubber and plastic products; basic metals and fabricated metal products;other non-metallic mineral products

29 Machinery and equipment n.e.c.31 Electrical machinery and apparatus n.e.c.34 Motor vehicles, trailers and semi-trailers35 Other transport equipment, excluding 35.3

Low-technology manufacturing (LT)

15 to 22 Food products, beverages and tobacco; textiles and textile products;leather and leather products; wood and wood products; pulp, paper and

paper products, publishing and printing

36 to 37 Manufacturing n.e.c.

SERVICES

Knowledge-intensive services (KIS)

61 Water transport62 Air transport64 Post and telecommunications65 to 67 Financial intermediation70 to 74 Real estate; renting and business activities

Other services (OS)

50 to 52 Wholesale; retail; motor trade60 Land transport, transport via pipelines63 Supporting and auxiliary tranport activities, activities of travel agencies90 Sewage and refuse disposal, sanitation and similar activities

Note: Data for hotel and restaurants (55), financial intermediation (65 to 67), public administration and defence, compulsory

social security (75), education (80), health and social work (85), activities of membership organization n.e.c. (91), recreational,

cultural and sporting activities (92), other service activities (93), activities of households (95 to 97) and extra-territorial

organizations and bodies (99) are not available.

56

Appendix B : Statistical annex

Table B.1: Estimation sample by country and 21-industry

GERMANY THE NETHERLANDS# obs. % # �rms % # obs. % # �rms %

Food 493 4.2 298 4.5 1,421 5.8 810 5.5

Textile 365 3.1 198 3.0 405 1.6 226 1.5

Wood 715 6.1 411 6.2 311 1.3 180 1.2

Chemicals 543 4.6 316 4.8 909 3.7 425 2.9

Plastics 528 4.5 292 4.4 590 2.4 295 2.0

Glass 357 3.1 205 3.1 466 1.9 264 1.8

Metal 1,124 9.6 596 9.0 1,926 7.8 1,134 7.6

Machinery 973 8.3 539 8.1 1,532 6.2 855 5.8

Electrical engineering 1,349 11.5 761 11.5 829 3.4 471 3.2

Vehicles 402 3.4 236 3.6 555 2.3 324 2.2

Furniture/recycling 337 2.9 196 3.0 528 2.1 332 2.2

Wholesale 468 4.0 243 3.7 4,624 18.8 2,841 19.1

Transport 801 6.8 448 6.8 2,954 12.0 1,744 11.8

Telecomm. 64 0.5 37 0.6 116 0.5 74 0.5

Computer 500 4.3 309 4.7 1,067 4.3 770 5.2

Technical services 712 6.1 388 5.8 964 3.9 611 4.1

Consultancy 394 3.4 253 3.8 1,360 5.5 933 6.3

Other business related serv. 808 6.9 503 7.6 2,699 11.0 1,697 11.4

Renting 237 2.0 112 1.7 218 0.9 143 1.0

Retail 282 2.4 137 2.1 1,056 4.3 668 4.5

RD services 247 2.1 156 2.4 56 0.2 44 0.3

Total 11,699 100.0 6,634 100.0 24,586 100.0 14,841 100.0

57

Table B.2: Panel structure: Number of participations

GERMANY THE NETHERLANDS# of participation # obs. % # �rms % # obs. % # �rms %

1 3,582 30.6 3,582 54.0 9,177 37.3 9,177 61.8

2 3,446 29.5 1,723 26.0 6,130 24.9 3,065 20.7

3 2,391 20.4 797 12.0 4,395 17.9 1,465 9.9

4 1,520 13.0 380 5.7 3,144 12.8 786 5.3

5 760 6.5 152 2.3 1,740 7.1 348 2.3

Total 11,699 100.0 6,634 100.0 24,586 100.0 14,841 100.0

58

Table B.3: Firm-level persistence in the closeness to the technological frontier, based on TFP (transition rates)

GERMANY THE NETHERLANDS<q20 q20 - q40 - q60 - q80 - ≥q95 <q20 q20 - q40 - q60 - q80 - ≥q95

<q40 <q60 <q80 <q95 <q40 <q60 <q80 <q95Populationa)

CTFTFPt CTFTFPt+1

<q20 81.15 13.81 2.95 1.30 0.59 0.19 78.11 14.86 4.17 1.93 0.72 0.21

q20 - <q40 13.88 66.52 14.85 3.56 1.05 0.13 11.95 66.06 16.08 4.46 1.23 0.23

q40 - <q60 2.54 15.48 64.24 15.17 2.31 0.27 3.30 14.73 63.04 15.65 2.80 0.49

q60 - <q80 1.03 3.71 16.04 66.08 12.31 0.84 1.57 4.34 14.03 67.12 11.95 0.99

q80 - <q95 0.76 1.82 3.44 16.55 71.30 6.13 0.80 1.49 3.50 15.85 72.34 6.02

≥q95 (frontier) 0.60 1.42 1.95 3.60 17.47 74.96 0.68 0.99 1.27 3.51 17.16 76.38CTFTFPt CTFTFPt+2

<q20 70.57 19.74 5.32 2.93 1.06 0.38 58.54 24.85 9.20 4.96 1.82 0.64

q20 - <q40 20.26 51.19 19.92 6.37 1.74 0.51 20.45 42.19 24.57 9.24 2.94 0.61

q40 - <q60 4.61 21.97 48.81 19.81 4.15 0.64 7.59 22.96 38.72 23.64 5.96 1.13

q60 - <q80 1.98 6.14 22.12 51.14 17.03 1.60 3.77 8.92 21.34 45.45 18.30 2.22

q80 - <q95 1.48 3.68 6.34 23.53 56.08 8.89 1.93 3.56 7.37 24.88 52.62 9.63

≥q95 (frontier) 1.15 3.45 4.44 6.58 28.45 55.92 1.86 2.04 3.25 7.87 28.88 56.09Estimation sampleb)

CTFTFPt CTFTFPt+2

<q20 66.58 20.99 6.68 4.01 1.47 0.27 60.36 24.26 9.04 3.89 1.69 0.76

q20 - <q40 22.14 46.29 20.50 8.68 1.76 0.63 21.50 40.00 25.79 9.57 2.50 0.64

q40 - <q60 4.34 23.63 45.59 20.56 5.49 0.38 6.68 24.59 38.87 24.03 4.84 0.98

q60 - <q80 2.53 5.81 24.49 48.74 16.67 1.77 3.70 9.12 23.84 44.12 16.51 2.71

q80 - <q95 2.05 5.60 7.65 21.64 52.61 10.45 2.15 3.83 7.75 25.68 49.49 11.10

≥q95 (frontier) 1.09 5.43 4.35 5.43 27.17 56.52 1.79 1.46 3.90 11.71 30.89 50.24

Notes: CTFTFP is divided into six categories based on the annual 20th, 40th, 60th, 80th and 95th percentiles. a) DE: 33,869 observations, NL: 238,259. observations.b) DE: 10,928 observations, NL: 24,591 observations.

59

Table B.4: Mean regression (OLS and FE): Firm-level returns to human capital and innovation, by country and industry


HT MT LT KIS OS HT MT LT KIS OSOLS HC -0.013 0.121* 0.554*** -0.010 0.343** 0.110*** 0.001 0.379*** 0.284** 0.349*** 0.726*** 0.569***

(0.067) (0.070) (0.130) (0.052) (0.153) (0.032) (0.075) (0.044) (0.132) (0.043) (0.050) (0.035)

L1.CTF 0.894*** 0.823*** 0.866*** 1.478*** 1.399*** 1.280*** 0.540*** 0.330*** 0.239*** 1.251*** 1.293*** 1.068***

(0.101) (0.057) (0.106) (0.092) (0.087) (0.038) (0.095) (0.036) (0.056) (0.066) (0.048) (0.027)

PD 0.017*** 0.017*** 0.026*** 0.039*** 0.069*** 0.029*** 0.015** 0.004** 0.014*** 0.043*** 0.053*** 0.036***

(0.005) (0.003) (0.006) (0.006) (0.007) (0.002) (0.006) (0.002) (0.004) (0.004) (0.004) (0.002)

PC -0.033 -0.013 -0.039 -0.095*** -0.101** -0.036*** -0.058* -0.036*** -0.031 -0.092*** -0.092*** -0.075***

(0.026) (0.016) (0.025) (0.030) (0.041) (0.011) (0.031) (0.009) (0.020) (0.023) (0.023) (0.009)

FE HC -0.082 -0.032 0.160 0.024 0.189 0.041 0.202 0.174*** -0.217 -0.022 0.038 0.053

(0.121) (0.085) (0.136) (0.062) (0.199) (0.042) (0.181) (0.057) (0.153) (0.049) (0.068) (0.035)

L1.CTF 0.537*** 0.497*** 0.256*** 0.884*** 0.680*** 0.567*** 0.190** 0.214*** 0.223*** 0.623*** 0.545*** 0.443***

(0.116) (0.052) (0.058) (0.144) (0.121) (0.046) (0.082) (0.036) (0.081) (0.064) (0.054) (0.026)

PD 0.007 0.017*** 0.011*** 0.029*** 0.023*** 0.020*** 0.015** 0.008*** 0.018*** 0.023*** 0.028*** 0.022***

(0.007) (0.003) (0.004) (0.006) (0.007) (0.002) (0.007) (0.002) (0.004) (0.003) (0.003) (0.002)

PC -0.010 -0.036*** -0.007 -0.012 -0.043* -0.023*** -0.032 -0.002 -0.001 -0.018 -0.049*** -0.031***

(0.025) (0.012) (0.015) (0.020) (0.026) (0.009) (0.037) (0.008) (0.017) (0.014) (0.012) (0.007)

Notes: Sample: firms with 2 or more observations. SIZE, CAP , MAT , GP , EAST (for DE), time dummies and industry dummies (total sample) have been includedin both OLS and FE regressions but are not reported here. Number of observations: See Table 1.

60

Table B.5: Quantile regression (QR): Firm-level returns to human capital and innovation, by country and industry


HC q10 -0.157 -0.140** 0.100 0.282*** 0.109 -0.025 0.045 0.175*** 0.096 0.651*** 0.715*** 0.326***(0.102) (0.067) (0.077) (0.078) (0.132) (0.041) (0.075) (0.027) (0.068) (0.057) (0.067) (0.033)

q25 0.029 -0.011 0.300*** 0.149*** 0.148 0.083*** 0.066 0.247*** 0.115** 0.594*** 0.692*** 0.433***(0.077) (0.049) (0.089) (0.047) (0.096) (0.029) (0.052) (0.025) (0.053) (0.046) (0.051) (0.031)

q50 0.080 0.119*** 0.547*** 0.041 0.329** 0.153*** -0.001 0.321*** 0.259*** 0.320*** 0.694*** 0.481***(0.051) (0.040) (0.127) (0.052) (0.133) (0.019) (0.056) (0.030) (0.065) (0.042) (0.037) (0.0283)

q75 0.120 0.140** 0.906*** -0.136** 0.484*** 0.170*** -0.106 0.352*** 0.311*** 0.040 0.704*** 0.507***(0.088) (0.070) (0.263) (0.067) (0.151) (0.036) (0.111) (0.043) (0.110) (0.039) (0.056) (0.029)

q90 0.093 0.334*** 0.533* -0.383*** 0.560* 0.139*** -0.114 0.545*** 0.360* -0.077 0.692*** 0.640***(0.141) (0.128) (0.280) (0.101) (0.306) (0.040) (0.125) (0.071) (0.188) (0.064) (0.078) (0.045)

L1.CTF q10 0.708*** 0.626*** 0.497*** 1.227*** 0.792*** 0.839*** 0.316** 0.112*** 0.080*** 0.790*** 1.0961*** 0.704***(0.140) (0.049) (0.049) (0.136) (0.099) (0.040) (0.115) (0.023) (0.028) (0.062) (0.052) (0.024)

q25 0.638*** 0.493*** 0.396*** 1.245*** 0.901*** 0.870*** 0.333*** 0.159*** 0.078*** 0.984*** 1.143*** 0.840***(0.080) (0.040) (0.025) (0.090) (0.084) (0.034) (0.079) (0.020) (0.017) (0.078) (0.038) (0.023)

q50 0.640*** 0.549*** 0.435*** 1.252*** 1.195*** 1.016*** 0.472*** 0.190*** 0.092*** 1.266*** 1.249*** 0.983***(0.081) (0.039) (0.057) (0.106) (0.064) (0.036) (0.093) (0.023) (0.022) (0.058) (0.032) (0.025)

q75 0.774*** 0.624*** 0.691*** 1.520*** 1.621*** 1.278*** 0.567*** 0.236*** 0.142*** 1.455*** 1.412*** 1.168***(0.079) (0.051) (0.105) (0.108) (0.102) (0.040) (0.129) (0.024) (0.036) (0.064) (0.046) (0.023)

q90 1.346*** 0.792*** 1.300*** 1.756*** 1.628*** 1.461*** 0.748*** 0.303*** 0.233*** 1.500*** 1.516*** 1.324***(0.206) (0.048) (0.100) (0.145) (0.143) (0.040) (0.112) (0.043) (0.073) (0.106) (0.057) (0.037)

PD q10 0.015** 0.013*** 0.005 0.039*** 0.050*** 0.019*** 0.010 0.006*** 0.007** 0.029*** 0.029*** 0.023***(0.007) (0.003) (0.003) (0.007) (0.009) (0.002) (0.007) (0.002) (0.003) (0.005) (0.005) (0.002)

q25 0.009** 0.010*** 0.009** 0.032*** 0.046*** 0.016*** 0.014*** 0.003*** 0.006*** 0.026*** 0.037*** 0.023***(0.005) (0.002) (0.004) (0.005) (0.006) (0.002) (0.005) (0.001) (0.002) (0.004) (0.003) (0.002)

q50 0.012** 0.010*** 0.012** 0.023*** 0.055*** 0.016*** 0.013** 0.003** 0.006*** 0.024*** 0.044*** 0.024***(0.005) (0.002) (0.004) (0.005) (0.009) (0.002) (0.006) (0.001) (0.002) (0.004) (0.004) (0.002)

q75 0.014** 0.015*** 0.023*** 0.031*** 0.068*** 0.023*** 0.018** 0.000 0.007* 0.035*** 0.059*** 0.029***(0.006) (0.003) (0.006) (0.004) (0.014) (0.003) (0.009) (0.002) (0.004) (0.006) (0.006) (0.002)

q90 0.007 0.021*** 0.044*** 0.053*** 0.106*** 0.034*** 0.020* -0.006 0.022*** 0.065*** 0.085*** 0.038***(0.012) (0.005) (0.015) (0.008) (0.024) (0.004) (0.012) (0.004) (0.009) (0.010) (0.010) (0.003)

PC q10 -0.031 -0.001 0.027 -0.128** -0.041 -0.006 -0.060 -0.013* 0.006 -0.084*** -0.031 -0.053***(0.028) (0.015) (0.022) (0.052) (0.058) (0.010) (0.041) (0.007) (0.012) (0.027) (0.021) (0.011)

q25 -0.005 -0.008 0.005 -0.072** -0.079** -0.011* -0.040 -0.019** 0.001 -0.075*** -0.042*** -0.048***(0.021) (0.009) (0.017) (0.031) (0.039) (0.006) (0.033) (0.008) (0.012) (0.023) (0.015) (0.008)

q50 0.005 -0.003 -0.019 -0.039 -0.086** -0.015* -0.047 -0.028*** -0.003 -0.035 -0.056*** -0.045***(0.022) (0.009) (0.015) (0.033) (0.034) (0.009) (0.029) (0.007) (0.012) (0.023) (0.017) (0.008)

q75 -0.040 -0.011 -0.044** -0.089** -0.057 -0.029*** -0.033 -0.031*** -0.026 -0.091*** -0.053** -0.059***(0.028) (0.012) (0.022) (0.036) (0.047) (0.011) (0.051) (0.009) (0.022) (0.025) (0.023) (0.010)

q90 -0.077 -0.006 -0.070 -0.156*** -0.102*** -0.041*** -0.028 -0.026 -0.069* -0.108*** -0.140*** -0.087***(0.064) (0.025) (0.047) (0.060) (0.038) (0.012) (0.055) (0.019) (0.042) (0.038) (0.025) (0.015)

Notes: Sample: firms with 2 or more observations (DE: 8,117; NL: 15,427 observations). Significance level of ***1%, **5%, *10%. Bootstrapped standard errors (20 replications). Quantile

regressions additionally include GP , EAST (for DE), time dummies and industry dummies (for total sample). Number of observations: See Table 1. Results are based on simultaneous regressions

for θ ∈ {0.05, 0.10, 0.20, 0.25, 0.30, 0.40, 0.50, 0.60, 0.70, 0.75, 0.80, 0.90, 0.95}. Results for other quantiles are available upon request.

61

Table B.6: Quantile regression (QR): Impact of human capital and innovation on industry productivity distribution, by country

GERMANY THE NETHERLANDS

HT MT LT KIS OS TOTAL HT MT LT KIS OS TOTAL

HC Dispersion 0.615 1.174 0.502*** -21.399 0.531** 0.345** 4.320 0.175*** 0.459** -0.873*** 0.009 0.079**

(0.738) (0.878) (0.127) (155.546) (0.249) (0.171) (11.68) (0.065) (0.192) (0.114) (0.046) (0.036)

Skewness -0.122 -0.715 0.184 0.244 -0.079 -0.613 0.225 -0.410 -0.470 0.011 0.622 -0.289

(1.418) (0.767) (0.282) (0.231) (0.476) (0.645) (0.530) (0.478) (0.736) (0.121) (5.531) (0.552)

Kurtosis -0.710 1.280 1.045** 0.352 1.993** 1.297 0.405 6.885*** 2.330** -1.037*** 115.4 13.02**

(2.512) (1.092) (0.514) (0.502) (1.002) (0.923) (0.817) (2.682) (1.253) (0.180) (609.3) (5.820)

L1.CTF Dispersion 0.096 0.117*** 0.271*** 0.100*** 0.285*** 0.190*** 0.260** 0.196*** 0.292** 0.193*** 0.108*** 0.163***

(0.060) (0.032) (0.068) (0.035) (0.038) (0.015) (0.115) (0.058) (0.124) (0.034) (0.018) (0.013)

Skewness 0.969 0.146 0.733** 0.946** 0.183 0.284** -0.185 0.197 0.552 -0.195 0.212 0.132

(1.050) (0.549) (0.309) (0.476) (0.189) (0.136) (0.553) (0.425) (0.529) (0.187) (0.180) (0.094)

Kurtosis 15.110* 10.876*** 6.096*** 10.812*** 3.364*** 5.645*** 4.549** 5.365*** 4.873** 4.854*** 9.721*** 6.180***

(8.604) (3.274) (1.872) (3.894) (0.528) (0.471) (2.210) (1.645) (2.459) (0.824) (1.639) (0.477)

PD Dispersion 0.187 0.194* 0.443*** -0.007 0.192** 0.178*** 0.133 -0.831 0.061 0.139 0.226*** 0.108***

(0.277) (0.101) (0.169) (0.078) (0.077) (0.058) (0.256) (1.041) (0.294) (0.096) (0.042) (0.036)

Skewness -0.434 1.056* 0.640 -37.244 0.150 1.058** 1.451 1.119 2.126 1.608 0.364 0.927*

(1.817) (0.562) (0.430) (408.621) (0.606) (0.438) (3.083) (0.935) (9.632) (1.239) (0.228) (0.559)

Kurtosis 4.995 7.091* 3.419* -200.523 7.114** 7.587*** 6.898 -0.012 35.00 11.00 5.285*** 10.87***

(7.763) (4.170) (1.822) (2155.764) (3.424) (2.663) (13.76) (1.612) (171.8) (7.846) (1.055) (3.688)

PC Dispersion 0.789 0.126 1.264 0.110 -0.159 0.460** -0.091 0.229 1.112 0.097 0.116 0.096

(0.763) (0.710) (1.038) (0.218) (0.367) (0.226) (0.693) (0.205) (1.048) (0.177) (0.236) (0.080)

Skewness 1.545 5.759 0.025 4.692 1.609 0.572 3.286 -0.578 0.677 5.896 -1.471 1.690

(0.989) (34.113) (0.535) (8.926) (3.613) (0.825) (22.47) (1.248) (0.822) (11.06) (4.392) (1.651)

Kurtosis 3.069 2.866 0.858 16.038 -6.598 2.546** 3.864 3.389 2.324 -13.19 11.96 13.56

(2.549) (22.889) (1.179) (31.654) (14.869) (1.220) (10.28) (3.209) (1.716) (96.64) (21.43) (11.46)

Notes: Sample: �rms with 2 or more observations (DE: 8,117; NL: 15,427 observations). Signi�cance level of ***1%, **5%, *10%. Tests are based on estimation

results in Table B.5.

62

Documents

Allocation of human capital and innovation at the …...Allocation of human capital and innovation at the frontier: Firm-level evidence on Germany and the Netherlands Eric BARTELSMANy